Metacognition Variability in Brain Structure Across
Measures and Domains in a University Population
Name: Marina Picó Cabiró
Student number: 12798770
Daily supervisor and internal assessor: Prof. Dr. Wouter van den Bos
Lab: Connected Minds Lab
Date of Submission: 01/10/2020
Word Count: 8.398
2
Abstract
Metacognition is the ability to reflect, monitor and control one’s own cognitive processes. While previous studies have centred on finding the neural substrates of metacognitive ability in the lab, they have overlooked what the neural correlates might be for other forms of metacognition like self-report measures. In this study, we brought together methods from the educational and neuroscientific fields and assessed domain-generality and specificity questions about the neural substrates of metacognition. Here, we found a dissociation between the measures of metacognitive ability, which suggests that there may be domain-specific networks for different domains of metacognition. Moreover, individual differences in grey matter volume in the rostral middle frontal cortex (rMFC) was associated with metacognitive regulation skills, and that variability in connectivity strength between caudal anterior cingulate cortex (cACC) and the precuneus was related to metacognitive ability in a perceptual task. Previous studies showed that the precuneus is mostly involved in metacognitive ability in memory tasks and not in perceptual tasks. Therefore, our results indicate that while the dissociation between domains is clearer at the behavioural level, metacognition is not only due to domain-specific neural processing signals, but also due to communication in this network. All in all, we suggest that metacognitive ability is likely to rely on the both domain-specific and -general levels of representations that coexist in the human brain across different domains and tasks. We argue that further interdisciplinary research that develops and clarifies a common ground between measures and domains of metacognitive ability is needed to extend these findings.
3
Metacognition Variability in Brain Structure Across Measures and Domains in a University Population
Metacognition is the ability to think about, monitor and control one’s own cognitive processes; in other words, metacognition is cognition about cognition (Flavell, 1987; Fleming & Dolan, 2012; Morales et al., 2018; Rouault, McWilliams, et al., 2018). Traditionally, this ability has been studied in the educational sciences, mostly using self-report studies (Dinsmore et al., 2008; Harrison & Vallin, 2018) and individual differences in this ability have been associated with academic success at different levels of schooling (Adey & Shayer, 1994; J. R. Baird, 1986; Vrugt & Oort, 2008; Young & Fry, 2008). Moreover, it has previously been shown that
metacognitive regulation skills are significantly higher for graduate as compared to
undergraduate students (Young & Fry, 2008). Also, it has been reported that metacognitive skills initially develop in separate domains, and later become generalized across domains (Veenman & Spaans, 2005). As such, researchers in the field of cognitive neuroscience and education sciences have commonly divided metacognition into two main domains: (1) knowledge about cognition and (2) regulation of cognition (Brown, 1987; Flavell, 1987). While the former encompasses awareness of one’s thought processes, the latter encompasses the planning and monitoring of these cognitive processes (Livingston, 2003; Schraw & Dennison, 1994). A frequently used self-report instrument that addresses both metacognitive knowledge and metacognitive control is the Metacognitive Awareness Inventory (MAI; (Harrison & Vallin, 2018). In order to assess the cognitive nature of metacognition both in daily life and in the laboratory, in this study we combine methods from educational sciences and neuroscience which have traditionally been studied independently.
4
As compared to self-report measures, metacognition has more recently been studied in the lab using experimental measures. Here, metacognition is often measured by relating levels of self-confidence in performance with actual performance on an experimental task (Fleming & Lau, 2014). This ability to discriminate between one’s good and bad performance is typically termed metacognitive sensitivity. Correspondingly, high levels of metacognitive sensitivity are noted when a high confidence rating accompanies correct responses and when low levels of confidence go with bad performance. Further, a very robust method to compute metacognitive sensitivity is the meta-d’, which is borrowed from signal detection theory (SDT; Galvin, Podd, Drga, & Whitmore, 2003). Nevertheless, the meta-d’ has been shown to be limited as it does not control for task difficulty or stimulus strength. One solution to this problem is using the meta-d’ framework developed by Maniscalco & Lau (2014) in which meta-d’ is relative to the level of difficulty. This measure, defined as metacognitive efficiency, allows researchers to compare scores of metacognition across individuals or task domains, and it is calculated by computing the ratio meta-d’/d’, where meta d’ is metacognitive sensitivity and d’ refers to how well a subject can discriminate between two signals (Fleming & Lau, 2014; Rouault, McWilliams, et al., 2018). In the past years, experimental studies have also helped with understanding the
neurocognitive mechanisms of metacognitive ability. It has been found that the dorsal anterior cingulate cortex/pre-supplementary motor area (dACC/pre-SMA) and anterior insular cortex (AIC) generalise across domains for metacognitive ability (Qiu et al., 2018), together with the ventromedial prefrontal cortex (vmPFC) and striatum (Morales et al., 2018). In addition, grey matter volume in the precuneus has been found to positively correlate with metacognitive efficiency in memory tasks, although not in perceptual tasks (McCurdy et al., 2013).
5
2018) has also been suggested to be involved in metacognition in the memory domain.
Conversely, variability in grey matter volume in the anterior prefrontal cortex (aPFC) (Fleming et al., 2014; McCurdy et al., 2013; Morales et al., 2018; Qiu et al., 2018; Rouault, McWilliams, et al., 2018; Vaccaro & Fleming, 2018) and white matter microstructure underlying the ACC (B. Baird et al., 2015) have been found to positively correlate with metacognitive efficiency for perception. Studies looking at functional connectivity strength suggest that connectivity between interoceptive cortices (cingulate and insula) and aPFC (B. Baird et al., 2013; Fleming & Dolan, 2012) may also establish a basis for underlying mechanisms of metacognition ability.
The results of all the aforementioned studies and brain areas suggest specific and general regions for metacognitive knowledge, albeit not for metacognitive regulation. However, these results are not very conclusive and reveal that the extent to which metacognition ability is tied to a specific cognitive process (e.g., memory or perception) or, else, if it is a cognitive skill that generalizes to different domains is poorly understood. Additionally, the ecological validity of laboratory measures of metacognition and to what extent these relate to metacognition as
measured in educational sciences, or in developmental psychology is not yet known. Finally, the neural correlates for individual differences on the MAI are not known either. In this study we will test the relationship between the different measures of metacognition to further develop our understanding of the underlying cognitive nature of this ability.
In the present project we aim to bring together different experimental and self-report measures of metacognition to investigate if metacognitive ability relies on either domain-general or domain-specific cognitive processes. To dive further into this question, we will focus on three separate yet related questions. First, we will examine the relation between the different measures of metacognition—the MAI and the meta-d’/d’—in two different domains. Here, we expect the
6
meta-d’/d’ in both tasks to be highly correlated to the MK scale of the MAI, as they are likely to
be measuring the same construct. Secondly, we will investigate the relationship between the different behavioural measures of metacognition and brain structure. Our aim is to examine if there is an overlap in the cognitive processes related to metacognitive skills in different domains and measures. In order to investigate this, we will base our ROIs selection on previous functional studies and aim to extend their findings by using methods that have been scarcely adopted in the metacognitive literature: grey matter volumes analyses and white matter tract strength analyses. The use of structural brain measures provides a valuable estimate of stable individual differences and can therefore help us elucidate the domain-specificity or global nature of the neural
correlates underlying metacognitive ability. We expect to find evidence for domain specificity networks in line with previous studies. We also expect that the individual differences in brain structures involved in metacognitive efficiency in the two behavioural tasks will overlap with the individual differences found for the MAI, specifically for the MK scale, as they measure
identical processes. Thirdly, we will examine the developmental differences in metacognition by looking at first- and third-year university students, expecting higher metacognitive scores for third-year students as compared to first-year students. Further, we expect that metacognitive skills in one domain will be more correlated with metacognitive skills in other domains for third-year students, suggesting a higher transference of knowledge across domains for older students.
Methods Participants
201 university students from the University of Amsterdam were recruited for the MAI online questionnaire. Structural images and diffusion MRI data were sampled for 156 of them.
7
The data from the rest were removed due to it being incomplete due to recruitment issues related to COVID-19 safety measures and consequent closure of our lab. The final sample included
N=156 participants (76 females and 80 males, between 17-55 years; M(SD)= 21.13 (4.36). 122 of
the participants were 1st year bachelor’s students and 34 of them were 3rd year bachelor’s
students. Some of the participants were international students and all had to be fluent in English to understand the instructions. Participants had normal or corrected to normal vision. Both left and right-handed participants were included. Each participant signed an informed consent for both the behavioural tasks and MRI scan, and the study was approved by ethical committee of the University of Amsterdam. Participants received monetary reward or research credits, in addition to a monetary bonus that was dependent on their performance.
Materials
Behavioural measures.
Self-report online questionnaires. We used the Metacognitive Awareness Inventory
(MAI; S6; Harrison & Vallin, 2018), which is a self-report questionnaire with 19 items extracted from the original 52-item version from Shraw & Dennison (1994). Each item was answered on a scale from 1 to 6 where 1 referred to ‘Not at all typical of me’ and 6 to ‘Very typical of me’. We used this reduced version because it has been shown to have a better fit and to be more adequate for between-group comparison than the original instrument (Harrison & Vallin, 2018). Based on the distinction between two major components of metacognition, the inventory consisted of two scales: metacognitive knowledge (MK) and metacognitive control (MC). In the MK scale there were items such as ‘I ask myself if I learned as much as I could have once I finish a task’, which assessed the awareness of one’s strengths and weaknesses, knowledge about strategies and why
8
and when to use them. The MC scale included items such as ‘I have control over how well I learn’ to measure abilities, such as planning, implementation, monitoring abilities. Participants completed the questionnaire online and were then invited to take part in the next stages of the study (behavioural tasks and MRI scanning).
Tasks. Participants performed two behavioural tasks in which they had to make a
2-alternative forced-choice (2AFC) about what they had memorized or perceived (depending on the task), followed by a confidence rating of their decision (Figure 1).
Perceptual task. A perceptual decision-making task was used together with trial-by-trial confidence ratings. After being presented by a fixation cross for 1000ms, participants were presented with different varying randomly positioned white dots inside two black boxes. This lasted for a total of 300ms. They had to judge which of two boxes contained a higher number of dots. Participants then provided their confidence level on a scale from 1 to 6, in which 1 stood for ‘certainly wrong’, and 6 ‘certainly correct’ (see Figure 1A). In total, participants performed 114 trials. Judgements of global confidence before and after the experiment was also asked (global pre-test and post-test confidence, respectively) (for further details see Rouault, Seow, Gillan, & Fleming, 2018). Despite using the same task as the one used in Rouault et al., (2018), we did not use the staircase approach implemented in their Experiment 2, as it has been shown that this manipulation is likely to inflate the metacognitive ability estimates (for a more detailed explanation, see Rahnev & Fleming, 2019).
Memory task. The memory task followed the same format as developed by Fleming, Ryu,
Golfinos, & Blackmon (2014) and it was programmed in MATLAB (MathWorks) using Psychtoolbox. Participants were presented with a list of 50 English words for 1 minute at the
9
Figure 1.
Behavioural tasks
Note: A. Perceptual decision-making task. Participants were asked to judge which of the boxes contained more dots
and to provide a confidence rating after each decision. There was no time limit for neither choice or confidence responses. Adapted from “Psychiatric Symptom Dimensions Are Associated With Dissociable Shifts in Metacognition but Not Task Performance” byRouault, M., Seow, T., Gillan, C. M., & Fleming, S. M., 2018,
Biological Psychiatry, 84(6), 443–451. https://doi.org/10.1016/J.BIOPSYCH.2017.12.017 B. Memory task.
Participants were asked to memorize a list of words for 1 minute, then to make a two-alternative-forced choice about which of the words they had seen in the previous list, and finally a confidence rating about their decision. Adapted from ‘Domain-specific impairment in metacognitive accuracy following anterior prefrontal lesions’, by Fleming, S. M., Ryu, J., Golfinos, J. G., & Blackmon, K. E. (2014). Brain, 137(10), 2811–2822.
https://doi.org/10.1093/brain/awu221 SWEAT CHAIR ROCK SOAP HEAD POLLEN BATTLE ARMOUR HORN WHEEL HAIL FOOTBALL ANIMAL BREAST TULIP WEAPON TUCK GLASS SWEAT + INFANT 10 seconds left… SWEAT CHAIR ROCK SOAP HEAD POLLEN BATTLE ARMOUR HORN WHEEL HAIL FOOTBALL ANIMAL BREAST TULIP WEAPON TUCK GLASS
List memorization (50 words) 10s left to memorize task (x50) (left, or right?) Stimuli and type I 2AFC Type II task: Confidence level
until response 3s
1 min (at the beginning of each block)
Time 1 2 3 4 5 6
Confidence?
10
beginning of each block and they were asked to memorize as many words as possible. Next, a series of 2AFC judgements of old/new words appeared on left and right of the screen (Figure 1B). ‘Old’ words were the ones from list that they had just memorised, and the rest were ‘new’. Participants were instructed to choose with the left or right arrow keys the word that they remembered seeing from the previous list (‘old’ word). After each trial participants had to indicate their confidence level on a scale from 1 (low confidence) to 6 (high confidence), and they were encouraged to use the whole scale. Each participant completed 200 memory trials divided into four blocks, that varied by the order of the words. The order of the blocks was counterbalanced between participants.
Exclusion criteria
Participants were excluded if accuracy was below chance (<55%) in either of the tasks, if they only used 3 or less scales from the rating confidence scales, and if there was a negative correlation between performance and confidence. Furthermore, individual trials were excluded if no answer was registered, and if response time (RT) was higher than 10s or if RT deviated by more than 3 standard deviations (SD) from the mean of that participant; with this criterion, up to 5% of the trials were removed. One participant was also excluded because of bad fit of the
meta-d’ to the confidence rating data, and two more because they had a negative meta-meta-d’/meta-d’
estimation, due to bad fit of their data. Two participants were excluded because of incomplete DWI data and three more because of data corruption. Using these criteria, across both
experiments and imaging data processing, a total of 22 of participants were excluded for further analysis, resulting in 134 participants (71 males and 63 females).
11 MRI data acquisition
The MRI scans were acquired on a Philips 3.0 Tesla Achieva scanner (Philips, Best, The Netherlands). A high-resolution structural scan of the head is made using 32 channel head coil and using parallel imaging (SENSE). A gradient echo pulse sequence was used including a repetition time (TR) of 11 ms, echo time (TE) of 5.2 ms, inversion time of 0.95 sec, field of view (FOV) of 256 x 256 x 180, matrix size of 368 x 318 x 257 slices, flip angle, and voxel size of 0.70 x 0.81 x 0.70 mm3 . The DWI are obtained using a voxel size of 2.00 x 2.00 x 2.00 mm3, TR of 2550 ms, TE of 77 ms, flip angle of 90º, FOV 224 x 224 x 140, and a matrix 112 x 112 x 60 slices.
MRI data preprocessing
The structural data were preprocessed using the automatic preprocessing pipeline (fMRIPrep version 1.5.4) provided by the Spinoza Centre for Neuroimaging (Esteban et al., 2019, 2020). This included artifact removal, cortical surface generation, skull-striping, cross-modal registration and standard space alignment.
The T1-weighted images were corrected for intensity non-uniformity with
N4BiasFieldCorrection (Tustison et al., 2010), and skull-striped with a Nipype implementation of the antsBrainExtraction.sh, which are both part of the Advanced Normalization Tools package (ANTs 2.2.0; Avants et al., 2011). Next, the brain tissue was segmented. Each participant’s structural image was segmented into white matter, grey matter and cerebrospinal fluid using FAST (FSL 5.0.9, fsl.fmrib.ox.ac.uk/, Zhang, Brady & Smith, 2001). Brain surfaces were reconstructed using recon-all (FreeSurfer 6.0.1., Dale, Fischl, & Sereno, 1999). Finally, volume-based spatial normalization was implemented and the T1w images were registered to the
12
standard MNI152NLin2009cAsym space with a nonlinear transformation computed by the antsRegistration (ANTs 2.2.0).
ROIs for T1w grey matter volumes
Volumes in each ROI of FreeSurfer's atlas were obtained directly from FreeSurfer's output
aseg.stats and aparc.stats files, which provide information about the segmentation done in both
hemispheres for different cortical (aparc) and subcortical (aseg) structures. The default parcellation uses the Desikan-Killiany atlas (Desikan et al., 2006). We further obtained a table which listed the main volumes in left and right hemispheres. The ROI that we selected for both hemispheres included the insula, precuneus, parahippocampal cortex (PHC), rostralmiddle frontal cortex (rMFC), and caudal ACC (in the DKT this is the same as dACC) (see Figure 2). To correct for head size variation across subjects the grey matter volume of each region was divided by the estimation of the total intracranial volume (eTIV) for each subject, which was also an output of the Freesurfer’s aparc.stats files. This is a broadly used method for head variability across subjects (L. M. O’Brien et al., 2011). All corrected grey matter volumes were
standardized into z-scores.
Non-parametric Spearman correlations were done for the grey matter volumes in the selected ROIs as normality assumptions were violated for left cACC, bilateral rMFC, precuneus and insula. Bilateral ROIs for which the correlations were higher than .66 were combined into one single predictor, and a new correlation table using Spearman’s method was run for the resulting ROIs (see Table S1).
13 Diffusion weighted data preprocessing
FSL preprocessing pipelines were followed, using the FDT (FMRIB's Diffusion Toolbox), which is a software tool for analysis of diffusion weighted images in FSL. These pipelines included four main steps that were run separately. Firstly, the topup tool was used to minimise distortion and susceptibility artifacts generated during DTI acquisition (Smith et al., 2004).
Figure 2
Selected ROIs for the extraction of grey matter volume and for the tractography analyses
Note: Hemispheres are In green Insula, in crema: rostralmiddlefrontal cortex, in red: caudal anterior
cingulate cortex. In blue: parahippocampal cortex. In purple: right anterior dorsolateral prefrontal cortex and left anterior dorsolateral prefrontal cortex. These two purple regions were the only ones we changed for the tractography analyses and were used instead of the rMFC. The rest were used for both grey matter and tractography analyses.
cACC precuneus rMFC insula PHC adlPFC dlPFC R
14
Secondly, the eddy correction tool was applied to correct for subject motion and eddy current distortions induced by changes in the magnetic field during the diffusion encoding (Andersson & Sotiropoulos, 2016). Thirdly, the DTI fit consisted of fitting a diffusion tensor at each voxel of the eddy corrected data. The output contained different files, including an anisotropy map and different eigenvalues (Madden, Bennett, & Song, 2009). Lastly, the BEDPOSTx (Bayesian Estimation of Diffusion Parameters Obtained using Sampling Techniques) was run in order to determine the number of crossing fibres per voxel with Markov Chain Monte Carlo sampling. Thus, it estimated the proportion of the diffusion signal that could be accounted for by the first fibre orientation (Behrens et al., 2003; 2007). This step created the necessary files for running the probabilistic tractography (note that tractography refers to ‘any method for estimating the
trajectories of the fibre tracts in the white matter’; O’Donnell & Westin, 2011). After each step, visual inspection was done using FSLeyes (McCarthy, 2020).
Additionally, some intermediate files (transformation matrices) were also produced for further ROI transformations across different spaces. To do this, a transformation from standard to T1w space and another one from T1w to diffusion space were conducted.
ROIs selection methods for tractography analyses
We created the ROIs in Table S2 by using FreeSurfer’s ROI atlas in FSL, and following the Desikan-killiany-Tourville (DKT) human brain cortical labelling protocol (Klein &
Tourville, 2012). We used this method because it estimates the ROIs in each individual brain and it therefore accounts for individual differences in folding patterns (i.e. differences in gyri and sulci), which allows a better precision in registering the anatomical features of the brain across subjects (Desikan et al., 2006).
15
The ROIs were extracted from the DKT atlas files obtained from FreeSurfer’s output (i.e.
desc-aparcaseg_dseg files), as masks. Then, using the transformation matrix obtained after
BEDPOSTx, the ROIs were non-linearly transformed from native space to MNI standard space and binarized again (with a threshold of 0.2) due to the transformation, which had resulted in a non-binary smoothed mask.
In order to obtain a more accurate ROI than the one provided by the DKT for the rMFC, we combined the two following methods. Based on Vaccaro & Fleming (2018), two 12mm spheres were created on MNI coordinates -50, 24, 28 (for the left dorsolateral prefrontal cortex; L dlPFC) and on MNI coordinates 28,50,26 (for right anterior dorsolateral prefrontal cortex; R adlPFC) (see Figure 2 and Table S2), which had found to be peak voxel activations in that study. Then, they were multiplied with the rMPFC ROI on each hemisphere so we would obtain only the overlap between the sphere and the region obtained in Freesurfer. All the ROIs were transformed to MNI space so they would all be in the same space.
White matter probabilistic tractography
Probabilistic tractography was run in the FSL software to trace the fibre bundles. To investigate the networks involved in the metacognitive ability, both at a general-domain level and at a specific-domain level, a total of 20 major tracts were traced for each participant (Figure 3; Table S3).
ROIs that were defined in the standard MNI space were registered back to the individual cortical surfaces in diffusion space. From the pair of ROIs, one was first used as the seed area and the other ROI (in both the same and the other hemisphere) was used as target, and the other way around. From each voxel of the seed ROI, 5000 streamlines were sent out, with 2000 steps
16
of 0.5mm of step length. The curvature threshold was 0.2. The propagation would stop when meeting voxels with FA values < 0.1. For the resulting streamlines a threshold was set at 1 streamline to remove some noise. Next, the fibres between pairs of ROIs were traced in both directions, binarized and multiplied. The parts that were overlapping in both directions were considered to represent the existing fibres that linked the ROIs. This strategy allowed to have a good balance between noise reduction for stronger connections and maintenance of information for the weaker ones (Liu et al., 2020). Further, a mask was built for each pathway linking the pairs of ROIs, with all the participants’ tracks included in it. A threshold of 50% of the
participants (i.e. 77) was applied to them in order to construct the final mask. The pathways for each individual were finally refined using these masks as a reference, so that they would only include the tracts where at least 50% of the participants had overlapping voxels.
Lastly, the white matter integrity was examined in each individual’s resulting tracts. Fractional anisotropy (FA), was computed at voxel-level and then averaged for each fibre. FA is a widely used normalized parameter that describes the degree of the restriction of water
molecules diffusion in different directions. Therefore, with a range between 0 and 1, larger FA values indicate higher white matter integrity, as there is a greater degree of restriction in
directional diffusivity (Madden et al., 2009), which indicates the presence of white matter fibres. The pathways cACC – PHC and R adlPFC/ L dlPFC – PHC for both left and right hemisphere were deleted because all FA values were 0, and therefore there was no physical tract. See Figure 3 for a visual summary of all the remaining pathways selected.
A correlation table for all the selected tracts was carried out to check for highly correlated tracts in both hemispheres so they could be merged into one single predictor. High correlations (r>.67) between bilateral cACC – precuneus, cACC – insula, insula – precuneus and PHC –
17
Figure 3.
Selected Pathways that were included in the multiple regression analyses.
cACC – insula pathway cACC – precuneus pathway
cACC – R adlPFC/L dlPFC pathway Precuneus – PHC pathway
Insula – Precuneus pathway Insula – PHC pathway
Insula – L dlPFC/R adlPFC R adlPFC/L dlPFC – precuneus pathway
Note: In blue: ROIs. In yellow: mask for 10% of participants. In red: mask representing the overlap of 50% of
participants. The pathways cACC – PHC and R adlPFC/ L dlPFC – PHC for both left and right hemisphere were deleted because all (or almost all) FA values were 0, and therefore there was no tract present. Furthermore, the tracts cACC – precuneus, cACC – insula, insula – precuneus and PHC – precuneus pathways from both hemispheres were merged for further analysis due to high correlations between them. dlPFC=dorsolateral prefrontal cortex; PHC= parahippocampal cortex; cACC= caudal anterior cingulate cortex.
18
precuneus pathways led us to decide to merge them. Despite their high correlation, L insula – L dlPFC and R insula – R adlPFC were not merged because they are not equivalent pathways -we had dlPFC in the left hemisphere and the adlPFC in the right one, which positions them in slightly different coordinates. Another correlation table was run for the resulting combinations (see Table S3).
Quantifying confidence Meta-d’
Metacognitive efficiency was measured with the behavioural tasks. To calculate
metacognitive efficiency scores, we estimated log (meta-d’/d’), which derives from SDT and has an enhanced statistical power compared to other measures (Fleming & Lau, 2014). Here, meta-d’ refers to the ability to discriminate between one’s good and bad performance (metacognitive sensitivity, or Type 2 sensitivity in SDT) and d’ the ability to discriminate between signals (Type 1 sensitivity in SDT) (Maniscalco & Lau, 2014). This way, meta-d’/d’ is a relative measure, as it compares metacognitive sensitivity to the level of processing capacity (d’). Hence, a value of 1 in d’/d’ represents a theoretical ideal value of metacognitive efficiency. Values <1 in meta-d’/d’ show that metacognition is lower than expected based on this model; a value of 0.8, for example, indicates 80% of metacognitive efficiency, where 20% of the sensory information available for the decision is lost when the metacognitive judgement was made (Morales et al., 2018). Values are theoretically capped below 0 and above 1. However, the use of an uncapped maximum likelihood estimation procedure may lead to negative values due to estimation error (Fleming et al., 2014). Furthermore, values above 1 can occur if confidence judgements have higher fidelity beliefs about the decision than the sensory information for the decision itself (Fleming & Daw, 2017). In these cases, it may be that subjects respond with partial information than necessary for the decision. Therefore, higher or lower values than d’ indicate that
19
metacognition is better or worse, respectively, than expected given task performance. This has been suggested to occur because first-order decisions and confidence might be separated and might be partly parallel processing streams (Fleming & Daw, 2017). We used the freely available
Hmeta-d toolbox (https://github.com/metacoglab/HMeta-d) developed by Fleming (2017) to fit the meta-d’ to confidence rating data and to calculate the metacognitive efficiency (meta-d’/d’
ratio).
Statistical Analyses
Firstly, we analysed all behavioural data from our 4 measures of metacognition. In order to examine the relationship between the meta-d’/d’ in the memory and the perceptual task, and the MAI subscales some correlations were computed between the different measures. Next, we investigated differences between 1st and 3rd year students both in the MC and MK scales scores
of the MAI, and for the meta-d’/d’ in the memory and in the perceptual tasks. To examine this, independent samples t-tests were carried out to compare each measure between 1st and 3rd year
students. In order to examine whether metacognitive ability is more transferable to different domains for 3rd year students than for 1st year students, correlations between measures were
computed for the two groups, separately. All statistical analyses were computed using R studio Version 1.3.959 (RStudio Team, 2020). Next, to investigate the relationship between brain structure and these different metacognition measures (MK, MC, d’/d’ in memory and
meta-d’/d’ in perception) 8 different multiple linear regression models were carried. On the one hand,
four of the models explored the relationship between grey matter volumes from our ROIs and the different measures. A multiple linear regression was computed to predict scores in the MK scale of the MAI based on study year, left cACC, right cACC, bilateral rMFC, bilateral insula,
20
with the same dependent variables to predict, this time, scores in the MC scale of the MAI. A third and fourth model were built with the same independent variables, to predict for meta-d’/d’ in the perceptual task and for meta-d’/d’ in the memory task, independently. On the other hand, four models were computed to explore the relationship between each of the measures and the structural connectivity strength of the following pathways: (1) L cACC – L dlPFC, (2) R cACC – R adlPFC, (3) L insula – L PHC, (4) R insula – PHC, (5) L precuneus – L dlPFC, (6) R
precuneus – R adlPFC, (7) L insula – dlPFC, (8) R insula – R adlPFC, (9) cACC – precuneus, (10) cACC – insula, (11) PHC – precuneus, (12) insula – precuneus. Two multiple linear regression models were computed to predict for MC and MK scores, respectively, with the 12 aforementioned pathways, controlling for sex, and adding study year as an extra predictor. Finally, two further multiple linear regressions were calculated with the same independent and demographic variables, where meta-d’/d’ was set as dependent variables for the memory task and for the perceptual task, correspondingly. In all the models, sex was coded with 0 for female and 1 for male, and study year with 0 for ‘Year 1’ and 1 for ‘Year 3’.
Results Behavioural measures
First, metacognitive ability across domains and measures was examined. To study the relationship between and within the different scores in the MAI scales, and metacognitive efficiency in the memory and perceptual task, a correlation table was made (see Table 1). As expected, we found a strong significant association between the metacognitive control and metacognitive knowledge scales in the MAI (rs=.65, p < .001). No significant association was found between any of the metacognitive scales of the MAI and the metacognitive efficiency
21
scores in the memory or the perceptual domain. We did not find a correlation between subjects’ metacognitive scores in the memory and the perceptual domain (r(132) = 0.056; p = 0.52).
Interestingly, metacognition efficiency was relatively low in the perceptual task (M=.67, SD=.35) compared to metacognitive efficiency in the memory task (M=1.10, SD=.30).
Furthermore, measures of metacognition did not differ between year of study (see Table S4, Figure S1). Correlations between domains and measures were examined for 1st and 3rd year
students, respectively (see Table S5). Both correlations did not differ significantly. Significance between the correlations was checked using the cocor package in R. Fisher z ratio for
correlations between the measures was not significant.
Table 1
Correlation between MAI scales and Meta-d’/d’
Metacognitive Knowledge (MAI)
Metacognitive
Control (MAI) efficiency memoryMetacognitive Mean and SD Metacognitive Knowledge (MAI) 3.48 (.62) Metacognitive Control (MAI) .671** 3.41 (.59) Metacognitive efficiency memory .024 .102 1.10 (.30) Metadecision efficiency perception .05 .033 .056 .67 (.35) Note: **p<0.01
22
Individual differences in brain structure and metacognitive ability Grey matter volume analyses
First, we investigated individual differences in grey matter volumes in relation to the
MAI. A multiple linear regression was carried to predict scores in the MK scale of the MAI, based on study year, left and right caudal ACC, bilateral rMFC, bilateral insula, bilateral
precuneus and bilateral PHC (see Table 2). We controlled for age and gender. The data met the assumptions of independence, homoscedasticity (equal variance), linearity and normality of residuals. No predictors were found to be significant. An equivalent model was computed for the MC scale scores (see Table 2). Assumptions of independence, homoscedasticity, linearity and normality were also satisfied. The rMFC grey matter volume was found to be a significant negative predictor of MC when controlling for the rest of the independent variables in the model (β=-0.335, p=0.0182; Figure 5A). This negative association suggests that for smaller volumes in the rMFC, the metacognitive control is higher. None of the other predictors were found to be significant. Tests to see if the data met the assumption of collinearity (correlation between the predictors) indicated that multicollinearity was not a concern in any of the models; the variance inflator factor (VIF) for each predictor was smaller than 2.7 (R. M. O’Brien, 2007).
Second, we examined individual differences in grey matter volumes in relation to
metacognitive efficiency in the two behavioural tasks. Two other models were built for the same
predictors to predict for metacognitive efficiency both in the memory and in the perceptual task (see Table 2). No significant relationships were found.
23
Table 2.
Statistical significance of each ROI volume in each model
Dependent variable:
MAI Meta-d’/d’ ratio
Metacognitive
knowledge (1) Metacognitive control (2) Memory (3) Perception (4) Gender - Male 0.131 (0.188) -0.121 (0.186) -0.051 (0.057) 0.026 (0.068) Study year 3 0.417 (0.218) 0.121 (0.216) 0.128 (0.066) 0.044 (0.078) L cACC 0.064 (0.106) 0.051 (0.104) 0.026 (0.032) 0.030 (0.038) R cACC -0.046 (0.100) 0.002 (0.098) 0.007 (0.030) -0.031 (0.036) rMFC -0.163 (0.142) -0.335* (0.140) -0.036 (0.043) -0.017 (0.051) Precuneus 0.106 (0.136) 0.052 (0.134) 0.032 (0.041) 0.017 (0.049) Insula 0.031 (0.147) 0.144 (0.145) 0.003 (0.044) -0.026 (0.053) PHC -0.006 (0.128) 0.104 (0.126) -0.032 (0.039) -0.013 (0.046) Constant -0.160 (0.143) 0.038 (0.141) 1.101** (0.043) 0.647** (0.051) AIC 393.1 389.7 48.5 311.9 Observations 134 134 134 134 R2 0.045 0.069 0.041 0.031 Adjusted R2 -0.016 0.010 -0.021 -0.031 Residual Std. Error (df = 125) 1.008 0.995 0.279 0.744 Note: *p<0.05**p<0.01
24 Connectivity strength analyses.
Next, the relationship between the connectivity strength of the selected tracts (see Figure 3) and MK, as measured in the MAI, was examined using a multiple linear regression. The independent variables in this model included the structural connectivity strength of the pathways in Figure 3, study year and gender (the full list of predictors can be found in Table 3). Another model was built with the same independent variables to predict for MC, as measured in the MAI. Assumptions of independence, homoscedasticity, linearity and normality were met for both models. Overall, we did not find any significant predictors for MK or for MC. Finally, the relationship between individual differences in tract strength between our ROIs and metacognitive efficiency in perceptual and mnemonic judgements was examined. Two other models were computed with the same predictors and with the outcome variables meta-d’/d’ in the memory task and meta-d’/d’ in the perceptual task, respectively. We found a significant relationship between cACC – precuneus connectivity strength and metacognitive efficiency in the perceptual task (β=0.251, p<0.01). Visualization for the correlation between them can be found in Figure 5B. Multicollinearity was not a concern for none of the models; VIFs for all predictors were smaller than 5.3 (R. M. O’Brien, 2007).
Discussion
In this study we set out to combine self-report measures and a memory and perceptual experimental task to asses metacognition across and within domains and measures. We explored the relationship between these measures and whether metacognitive ability in the experimental tasks and in the MAI rely on the same neural correlates. Additionally, we explored
25
Table 3.
Statistical significance of each path in each model
Dependent variable
MAI Meta-d’/d’ ratio
Metacognitive
knowledge Metacognitive control Memory Perception
(1) (2) (3) (4) Study year 3 0.363 (0.234) 0.001 (0.235) 0.083 (0.065) 0.029 (0.166) Gender - Male 0.113 (0.214) -0.145 (0.215) -0.052 (0.060) 0.070 (0.152) L cACC – dlPFC (FA) 0.088 (0.105) 0.051 (0.105) 0.017 (0.029) -0.045 (0.074) R cACC – dlPFC (FA) 0.063 (0.161) 0.131 (0.162) 0.013 (0.045) 0.143 (0.115) L insula – PHC (FA) -0.012 (0.102) -0.063 (0.102) 0.018 (0.028) 0.073 (0.072) R insula – PHC (FA) 0.116 (0.109) -0.021 (0.110) -0.005 (0.030) -0.123 (0.077) L precuneus – dlPFC (FA) -0.215 (0.126) 0.099 (0.127) -0.007 (0.035) -0.134 (0.089) R precuneus – dlPFC (FA) 0.007 (0.137) -0.197 (0.138) -0.043 (0.038) -0.045 (0.097) L insula – dlPFC (FA) -0.110 (0.153) -0.205 (0.154) -0.030 (0.043) 0.063 (0.108) R insula – dlPFC (FA) -0.210 (0.199) -0.104 (0.200) 0.003 (0.056) -0.215 (0.141) cACC – precuneus (FA) 0.134 (0.114) 0.167 (0.114) 0.002 (0.032) 0.251** (0.080)
cACC – insula (FA) 0.160 (0.169) -0.010 (0.170) 0.003 (0.047) -0.123 (0.120) PHC – precuneus (FA) -0.045 (0.109) -0.011 (0.109) -0.053 (0.030) 0.051 (0.077) Insula – precuneus (FA) 0.040 (0.134) 0.052 (0.135) 0.035 (0.037) 0.058 (0.095) Constant -0.138 (0.150) 0.077 (0.151) 0.069 (0.042) -0.638** (0.106) AIC 397.5 398.9 56 305.4 Observations 134 134 134 134 R2 0.097 0.088 0.072 0.156 Adjusted R2 -0.009 -0.019 -0.037 0.056 Residual Std. Error (df = 119) 1.004 1.010 0.281 0.712 Note: *p<0.5**p<0.01
26
and third-year university students. We found that the memory and perceptual tasks were independent from each other and also from the MAI, suggesting that different types of
metacognition rely on different mechanisms. Furthermore, we found that individual differences in grey matter volume in the rMFC predicts differences in MC. Individual differences in
connectivity strength between cACC and the precuneus was found to be a significant predictor of differences in metacognitive ability in the perceptual task. All findings regarding each of our questions are reported and explained in detail next.
As expected, we found that MAI subscales (MC and MK) highly correlated with each other. Additionally, we expected the MK scale to be correlated with metacognitive capacity in
Figure 5.
Visualization of significant predictors in linear regression models.
A. In the models including grey matter volumes as predictors, negative correlation between metacognitive
control scale in the MAI and rMFC grey matter volume was found. Volume in the rMFC is negatively correlated with the scores obtained in the MC scale of the MAI. B. Relevant correlations from significant
predictors in linear regression models of connectivity strength between selected pathways. Positive correlation between the connectivity strength in the pathway caudal ACC – precuneus and the metamemory ability.
A B
r = -1.512
27
perceptual and memory decisions, as they are likely supposed to measure the same construct. That is, metacognitive efficiency is based on retrospective knowledge of previous decisions, and it measures awareness of one’s cognitive processes, similar to how it is done with the MK. In light of this, we did not find an association between the MC and the MK scales of the MAI and metacognitive capacity of perceptual and memory decisions, which suggests that the self-report and experimental measures of metacognitive ability may be independent from the other and may be measuring different aspects of metacognition. The independence of the two measures could indicate an issue of construct validity with the lab measures, which mostly measure domain-specific components of metacognition, while self-report measures treat metacognition as a general skill. The lack of correlation between the different measures also suggests that
metacognitive ability assessed in each of the measures might not only differ in domain, but also in its nature. It may be that rating confidence on an experimental task and reflecting about and self-reporting our learning skills and capabilities underlie different cognitive processes. Some researchers claim that metacognition does not have to be conscious and two levels of
metacognition can be distinguished, one that is high-level and another that is low-level (Arango-Muñoz, 2011). Respectively, the high-level relies on a self-ascription of mental states that depends on a capacity to infer, interpret and reflect on one’s mental states. Low-level metacognition is guided by our emotional states; unconscious feelings of confidence or uncertainty are able to control one’s cognitive abilities, without any introspective effort by the subject. This distinction, made in theoretical cognitive neuroscience research, would intuitively comprise metacognitive efficiency measured in the memory and perceptual tasks, as well the MAI scales into the high-level metacognition, as they are conscious, controlled and rational self-interpretations of one’s cognitive processes. However, in educational science this distinction has
28
not typically been used and it is therefore unclear whether a self-report like the MAI can indeed be encompassed in this division. It would be relevant to explore what theoretical differentiations like this or others could bring to clarify the mechanisms underlying different measures or settings of metacognition in order to build bridges across them or else, to clarify their incompatibilities. All in all, the fact that lab measures address metacognition in domain-specific areas, makes it difficult to generalise findings to all kind of domains and real-life settings where metacognition is involved. Finally, we found no association between metacognition in a memory task and metacognition in a perceptual task. Moreover, metacognitive efficiency was significantly higher for the memory task than for the perceptual task, which concurs with previous studies (Morales et al., 2018). This dissociation in metacognitive efficiency across domains is in line with previous findings (B. Baird et al., 2013, 2015; Fitzgerald et al., 2017; Fleming et al., 2014; Morales et al., 2018) and it suggests that metacognitive ability underlies different mechanisms in different domains. One could ponder that this dissociation is somewhat inconsistent with the aforementioned assumption of MK being correlated to metacognitive efficiency in the memory and perceptual domain, which would imply a general-domain network. We propose that both expectations could converge in a framework where both domain general and specific
mechanisms coexist and communicate toward a metacognitive ability. This idea is developed further in the third point of the discussion.
Secondly, we explored developmental differences in metacognition between 1st and 3rd year
university students. Previous studies have shown differences among adult (Schraw, 1994) and young learners (Young & Fry, 2008) in relation to regulatory skills, but not in metacognitive knowledge. However, we found no differences in neither the MC or the MK scales of the MAI between 1st and 3rd year university students. The absence of differences in the MC scale between
29
the two groups was not expected. Further, no differences were found between 1 and 3 year bachelor’s students in metacognitive efficiency for the perceptual and for the memory task. Although developmental differences have not been yet explored in lab measures, we had
expected 3rd year students to showcase better performance than 1st year students in metacognitive
efficiency for both tasks, especially in the memory task, as they are likely to be more trained in this domain.
Moreover, recent theories propose that MK abilities are initially domain-specific and then they evolve to become gradually more domain-independent, as knowledge and practice is acquired and linked between domains. To study these developmental differences, we examined sets of correlations between the four tasks amongst 1st year students and 3rd year students. When
comparing the correlations between these two groups, no significant differences were found. These findings indicate that the correlations between the measures did not differ between groups and that, against expectations, older students did not show a higher transference of knowledge across tasks (as would be evident by higher correlations between the measures). This
transference of knowledge across domains has been previously studied in areas like text studying and problem solving, which are all common domains in educational sciences (Veenman et al., 2006). Yet, experimental domains (e.g. perception vs. memory tasks) might arguably be considered of a different nature, making it difficult for results to be generalized to all kinds of domains.
In school, situations (e.g., via teachers) provide for an environment that supports constant feedback for students in different educational tasks like writing, reading or problem solving. This way, students end up acquiring a general metacognitive ability that transfers to other schooling-related metacognitive domains. Following this line of thought we would have expected
30
improvements regarding metacognitive ability in the memory task, as memory is a strongly trained skill in academic life. Nonetheless, the difference in settings can also be accountable for this lack of developmental differences across domains. More importantly, the lack of consistency across measures might have been caused by the use of metacognitive efficiency (meta-d’/d’) as a measure of metacognitive ability in the experimental tasks, instead of metacognitive sensitivity (meta-d’). As previously mentioned in the methods, metacognitive efficiency controls for task performance, while metacognitive sensitivity (meta-d’) does not. This avoids making misguided abstractions about meta-d’ scores that might be affected by different cognitive processes (such as low-level perceptual processes) that are not per se metacognitive. Therefore, recent studies find meta-d’/d’ more ideal to compare metacognition across experimental domains. However, in our study we introduced the MAI in our measures and domains comparison, and this is a measure that does not control for differences in task performance (or in its respective context, for academic performance). Hence, meta-d’/d’ might be less adequate to do comparisons across settings and between different groups that might differ in skill. Due to time limitations the comparison between groups could not be checked using meta-d’; this is certainly an area worth further examination.
Another alternative solution would be controlling for performance on the MAI side, and examining the relationship between the MAI scores and academic performance (e.g. average grades). Lastly, a final possible reason of these results might be due to our unbalanced sample (N=109 in the 1st year-student group vs. N=29 in the 3rd year student one). Results should be
compared again when the two groups are more balanced. Using samples with bigger differences like high-school students vs. master’s students, or even better, carrying out longitudinal studies would be key in clarifying the developmental evolution of metacognition over the years in young
31
learners. Including students from different backgrounds could also provide more insights into how differences in learning abilities in real life settings relate to differences in domain-generality or specificity for metacognitive ability in the tasks we used in the lab.
Thirdly, we looked at the relationship between individual differences in brain structure in relation to differences in metacognition. We found that individual differences in rMFC grey matter volumes were correlated differently in metacognitive control. The rMFC that we selected in our ROIs is an area that we extracted using from the DK atlas (Desikan et al., 2006). This region contains subregions that have also been investigated in relation to metacognitive ability: the aPFC and dlPFC, which are also part of Broadman area (BA) 10 and BA46, respectively. Both the aPFC and dlPFC have been found to play a role in metacognitive ability in perceptual tasks (B. Baird et al., 2013; Fleming et al., 2012, 2014; Fleming & Dolan, 2012; Vaccaro & Fleming, 2018). Besides, previous findings have shown that greater activation of regions like the anterior anterior lateral prefrontal cortex (lPFC) is related to metacognitive regulation of decision adjustment (Qiu et al., 2018). Some authors have even suggested that the dlPFC activity
underlies the selection of metacognitive responses (Fleming & Dolan, 2012). Therefore, our finding confirms the involvement of the rMFC in metacognitive ability, although it remains unclear clear (1) whether that is specifically for a domain or type of metacognition (like MC), or instead, if the rMFC is involved in metacognition in a domain-general fashion, and (2) if the subregions included in the rMFC (aPFC and dlPFC) make differential contributions to
metacognitive ability. Examining metacognitive regulation not only in self-reports, but also in experimental tasks, like in a re-decision paradigm, such as the one used in Qiu et al. (2018) — where participants could revise their initial decision as well as confidence on that decision— would help elucidating if these findings generalize across paradigms for metacognitive control.
32
Overall, it is crucial to further investigate this relationship and specially the reason why the correlation between volume in the rMFC and MC is negative, when previous findings have found contrary findings. Such a study would benefit from focusing on more specific and clearly defined regions that are referred to with the same name, in order to avoid confusion and favour replicability,
To further examine our third research question about brain structure and metacognition, in this study, we also examined, the relationship between white matter connectivity strength and metacognitive ability. We found a significant relationship between white matter FA in the precuneus to cACC pathway and metacognitive efficiency in a perceptual task. Previous studies have shown that grey matter volumes in the precuneus were related to metacognitive efficiency in a memory task (McCurdy et al., 2013), but not in the perceptual task. Conversely, our results contradict this, as the precuneus to cACC pathway was significantly correlated to metacognitive efficiency in a perceptual task, but not in a memory task. Accordingly, this finding suggests that the precuneus is not specifically tied to mnemonic judgements, rather, it might play a role in metacognitive skills in the perceptual domain also. Hence, domain-specificity is not clear for metacognition and there might be at least some degree of crosstalk between domains (McCurdy et al., 2013; Rouault, McWilliams, et al., 2018).
Finally, no overlap was found between the measures and brain structure. We restricted our ROIs based on regions whose functional activity had previously been related to metacognitive ability, both at a domain-general and at a specific level. However, our findings did not fully verify the involvement of these regions in metacognitive ability when using structural methods like grey matter volume and connectivity strength analyses. This lack of validation shows that drawing consistent conclusions about the neural correlates underlying metacognition can be quite
33
challenging. A hypothesis-driven testing approach, like the one we took seems to be limited as the hypotheses are based on studies with small samples, which makes it difficult to generalise and confirm their results decisively. Further studies should aim at combining both exploratory and machine learning approaches, together with confirmatory analyses with bigger samples.
Altogether, our findings reveal that domain-generality is not as clear-cut for metacognition. On the one hand, at the behavioural level, we find a clear dissociation between domains (i.e. memory vs. perceptual) and measures (i.e. experimental vs. self-report). This however challenges the ecological and construct validity of lab measures, which study metacognition from a very domain-specific view and exclusively in experimental settings, without much regard for the commonalities with measures of metacognition that take place in real-life settings. On the other hand, we found that grey matter volume variability in the rMFC is associated with metacognitive regulation skills. Moreover, structural connectivity strength between the cACC and precuneus predict differences in metacognitive efficiency in a perceptual task. These findings evidence that domain specificity is not clear as a region like the precuneus was previously found to play a role in metacognition in memory tasks. We propose that while there are differences in the skills involved in metacognitive ability in different domains of metacognition, there might be an overlapping neural circuitry across domains and measures of metacognition. Further research should aim at bringing together different methods and disciplines that study metacognition in order to establish common ground between controlled experimental measures and real-life measures. Clarifying the underlying nature of metacognition regardless of the methods or setting where it is being studied can certainly provide relevant implications in real life contexts such as educational sciences, where learners are constantly reflecting on their cognitive abilities. This
34
could help improving both the way learners acquire and develop their learning abilities and the methods that are being used in teaching.
35
References
Adey, P., & Shayer, M. (1994). Really Raising Standards. Routledge.
https://books.google.com/books?hl=en&lr=&id=zyBO7qPErNsC&oi=fnd&pg=PT10&dq=shayer+a dey&ots=NAovaoU4ue&sig=VojkpapdFb8FQf9v8bB2f3jJ5fI#v=onepage&q=shayer adey&f=false
Andersson, J. L. R., & Sotiropoulos, S. N. (2016). An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. NeuroImage, 125, 1063–1078.
https://doi.org/10.1016/j.neuroimage.2015.10.019
Arango-Muñoz, S. (2011). Two Levels of Metacognition. Philosophia, 39(1), 71–82. https://doi.org/10.1007/s11406-010-9279-0
Avants, B. B., Tustison, N. J., Song, G., Cook, P. A., Klein, A., & Gee, J. C. (2011). A reproducible evaluation of ANTs similarity metric performance in brain image registration. NeuroImage, 54(3), 2033–2044. https://doi.org/10.1016/j.neuroimage.2010.09.025
Baird, B., Cieslak, M., Smallwood, J., Grafton, S. T., & Schooler, J. W. (2015). Regional white matter variation associated with domain-specific metacognitive accuracy. Journal of Cognitive
Neuroscience, 27(3), 440–452. https://doi.org/10.1162/jocn_a_00741
Baird, B., Smallwood, J., Gorgolewski, K. J., & Margulies, D. S. (2013). Medial and Lateral Networks in Anterior Prefrontal Cortex Support Metacognitive Ability for Memory and Perception. Journal of
Neuroscience, 33(42), 16657–16665. https://doi.org/10.1523/JNEUROSCI.0786-13.2013
Baird, J. R. (1986). Improving learning through enhanced metacognition: A classroom study. European
Journal of Science Education, 8(3), 263–282. https://doi.org/10.1080/0140528860080303
Behrens, T. E.J., Berg, H. J., Jbabdi, S., Rushworth, M. F. S., & Woolrich, M. W. (2007). Probabilistic diffusion tractography with multiple fibre orientations: What can we gain? NeuroImage, 34(1), 144–
36 155. https://doi.org/10.1016/j.neuroimage.2006.09.018
Behrens, Timothy E.J., Woolrich, M. W., Jenkinson, M., Johansen-Berg, H., Nunes, R. G., Clare, S., Matthews, P. M., Brady, J. M., & Smith, S. M. (2003). Characterization and Propagation of Uncertainty in Diffusion-Weighted MR Imaging. Magnetic Resonance in Medicine, 50(5), 1077– 1088. https://doi.org/10.1002/mrm.10609
Brown, A. L. (1987). Metacognition, executive control, self-regulation, and other more mysterious mechanisms. In Metacognition, motivation, and understanding (pp. 65–116). Lawrence Erlbaum Associates Publishers.
Dale, A. M., Fischl, B., & Sereno, M. I. (1999). Cortical surface-based analysis: I. Segmentation and surface reconstruction. NeuroImage, 9(2), 179–194. https://doi.org/10.1006/nimg.1998.0395
Desikan, R. S., Ségonne, F., Fischl, B., Quinn, B. T., Dickerson, B. C., Blacker, D., Buckner, R. L., Dale, A. M., Maguire, R. P., Hyman, B. T., Albert, M. S., & Killiany, R. J. (2006). An automated labeling
system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest.
https://doi.org/10.1016/j.neuroimage.2006.01.021
Dinsmore, D. L., Alexander, P. A., & Loughlin, S. M. (2008). Focusing the conceptual lens on
metacognition, self-regulation, and self-regulated learning. Educational Psychology Review, 20(4), 391–409. https://doi.org/10.1007/s10648-008-9083-6
Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., Kent, J. D., Goncalves, M., DuPre, E., Snyder, M., Oya, H., Ghosh, S. S., Wright, J., Durnez, J., Poldrack, R. A., & Gorgolewski, K. J. (2019). fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature
Methods, 16(1), 111–116. https://doi.org/10.1038/s41592-018-0235-4
37
R. A., & Gorgolewski, K. J. (2020). fMRIPrep: a robust preprocessing pipeline for functional MRI. https://doi.org/10.5281/ZENODO.3876458
Fitzgerald, L. M., Arvaneh, M., & Dockree, P. M. (2017). Domain-specific and domain-general processes underlying metacognitive judgments. Consciousness and Cognition, 49, 264–277.
https://doi.org/10.1016/J.CONCOG.2017.01.011
Flavell, J. H. (1987). Speculations about the nature and development of metacognition. In Metacognition
motivation and understanding (pp. 21–29). Lawrence Erlbaum Associates.
Fleming, S. M. (2017). HMeta-d: hierarchical Bayesian estimation of metacognitive efficiency from confidence ratings. Neuroscience of Consciousness, 2017(1). https://doi.org/10.1093/nc/nix007
Fleming, S. M., & Daw, N. D. (2017). Self-evaluation of decision-making: A general bayesian framework for metacognitive computation. Psychological Review, 124(1), 91–114.
https://doi.org/10.1037/rev0000045
Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical
Transactions of the Royal Society B: Biological Sciences, 367(1594), 1338–1349.
https://doi.org/10.1098/rstb.2011.0417
Fleming, S. M., Huijgen, J., & Dolan, R. J. (2012). Prefrontal contributions to metacognition in perceptual decision making. Journal of Neuroscience, 32(18), 6117–6125.
https://doi.org/10.1523/JNEUROSCI.6489-11.2012
Fleming, S. M., & Lau, H. C. (2014). How to measure metacognition. In Frontiers in Human
Neuroscience (Vol. 8, Issue JULY, p. 443). Frontiers. https://doi.org/10.3389/fnhum.2014.00443
Fleming, S. M., Ryu, J., Golfinos, J. G., & Blackmon, K. E. (2014). Domain-specific impairment in metacognitive accuracy following anterior prefrontal lesions. Brain, 137(10), 2811–2822.
38 https://doi.org/10.1093/brain/awu221
Galvin, S. J., Podd, J. V., Drga, V., & Whitmore, J. (2003). Type 2 tasks in the theory of signal detectability: Discrimination between correct and incorrect decisions. Psychonomic Bulletin &
Review, 10(4), 843–876. https://doi.org/10.3758/BF03196546
Harrison, G. M., & Vallin, L. M. (2018). Evaluating the metacognitive awareness inventory using empirical factor-structure evidence. Metacognition and Learning, 13(1), 15–38.
https://doi.org/10.1007/s11409-017-9176-z
Klein, A., & Tourville, J. (2012). 101 Labeled Brain Images and a Consistent Human Cortical Labeling Protocol. Frontiers in Neuroscience, 6(DEC), 171. https://doi.org/10.3389/fnins.2012.00171
Liu, X., Hildebrandt, A., Meyer, K., Sommer, W., & Zhou, C. (2020). Patterns of individual differences in fiber tract integrity of the face processing brain network support neurofunctional models.
NeuroImage, 204(September 2019), 116229. https://doi.org/10.1016/j.neuroimage.2019.116229
Livingston, J. A. (2003). Metacognition: An Overview.
Madden, D. J., Bennett, I. J., & Song, A. W. (2009). Cerebral white matter integrity and cognitive aging: Contributions from diffusion tensor imaging. In Neuropsychology Review (Vol. 19, Issue 4, pp. 415–435). Springer. https://doi.org/10.1007/s11065-009-9113-2
Maniscalco, B., & Lau, H. (2014). Signal Detection Theory Analysis of Type 1 and Type 2 Data: Meta-d′, Response-Specific Meta-d′, and the Unequal Variance SDT Model. In The Cognitive Neuroscience
of Metacognition (pp. 25–66). Springer Berlin Heidelberg.
https://doi.org/10.1007/978-3-642-45190-4_3
39
McCurdy, L. Y., Maniscalco, B., Metcalfe, J., Liu, K. Y., de Lange, F. P., & Lau, H. (2013). Anatomical Coupling between Distinct Metacognitive Systems for Memory and Visual Perception. Journal of
Neuroscience, 33(5), 1897–1906. https://doi.org/10.1523/JNEUROSCI.1890-12.2013
Morales, J., Lau, H., & Fleming, S. M. (2018). Domain-General and Domain-Specific Patterns of Activity Supporting Metacognition in Human Prefrontal Cortex. The Journal of Neuroscience : The Official
Journal of the Society for Neuroscience, 38(14), 3534–3546.
https://doi.org/10.1523/JNEUROSCI.2360-17.2018
O’Brien, L. M., Ziegler, D. A., Deutsch, C. K., Frazier, J. A., Herbert, M. R., & Locascio, J. J. (2011). Statistical adjustments for brain size in volumetric neuroimaging studies: Some practical
implications in methods. Psychiatry Research - Neuroimaging, 193(2), 113–122. https://doi.org/10.1016/j.pscychresns.2011.01.007
O’Brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality and
Quantity, 41(5), 673–690. https://doi.org/10.1007/s11135-006-9018-6
O’Donnell, L. J., & Westin, C. F. (2011). An introduction to diffusion tensor image analysis. In
Neurosurgery Clinics of North America (Vol. 22, Issue 2, pp. 185–196).
https://doi.org/10.1016/j.nec.2010.12.004
Qiu, L., Su, J., Ni, Y., Bai, Y., Zhang, X., Li, X., & Wan, X. (2018). The neural system of metacognition accompanying decision-making in the prefrontal cortex. PLOS Biology, 16(4), e2004037.
https://doi.org/10.1371/journal.pbio.2004037
Rahnev, D., & Fleming, S. M. (2019). How experimental procedures influence estimates of metacognitive ability. Neuroscience of Consciousness, 2019(1). https://doi.org/10.1093/nc/niz009
40
Domains: Insights from Individual Differences and Neuroimaging. Personality Neuroscience, 1, e17. https://doi.org/10.1017/pen.2018.16
Rouault, M., Seow, T., Gillan, C. M., & Fleming, S. M. (2018). Psychiatric Symptom Dimensions Are Associated With Dissociable Shifts in Metacognition but Not Task Performance. Biological
Psychiatry, 84(6), 443–451. https://doi.org/10.1016/J.BIOPSYCH.2017.12.017
RStudio Team. (2020). RStudio: Integrated Development Environment for R (1.3.959). RStudio, PBC. https://rstudio.com/
Schraw, G. (1994). The Effect of Metacognitive Knowledge on Local and Global Monitoring.
Contemporary Educational Psychology, 19(2), 143–154. https://doi.org/10.1006/ceps.1994.1013
Schraw, G., & Dennison, R. S. (1994). Assessing metacognitive awareness. In Contemporary
Educational Psychology (Vol. 19, Issue 4, pp. 460–475). https://doi.org/10.1006/ceps.1994.1033
Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E. J., Johansen-Berg, H., Bannister, P. R., De Luca, M., Drobnjak, I., Flitney, D. E., Niazy, R. K., Saunders, J., Vickers, J., Zhang, Y., De Stefano, N., Brady, J. M., & Matthews, P. M. (2004). Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage, 23(SUPPL. 1).
https://doi.org/10.1016/j.neuroimage.2004.07.051
Tustison, N. J., Avants, B. B., Cook, P. A., Zheng, Y., Egan, A., Yushkevich, P. A., & Gee, J. C. (2010). N4ITK: Improved N3 bias correction. IEEE Transactions on Medical Imaging, 29(6), 1310–1320. https://doi.org/10.1109/TMI.2010.2046908
Vaccaro, A. G., & Fleming, S. M. (2018). Thinking about thinking: A coordinate-based meta-analysis of neuroimaging studies of metacognitive judgements. Brain and Neuroscience Advances, 2,
41
Veenman, M. V. J., & Spaans, M. A. (2005). Relation between intellectual and metacognitive skills: Age and task differences. Learning and Individual Differences, 15(2), 159–176.
https://doi.org/10.1016/j.lindif.2004.12.001
Veenman, M. V. J., Van Hout-Wolters, B. H. A. M., & Afflerbach, P. (2006). Metacognition and learning: Conceptual and methodological considerations. In Metacognition and Learning (Vol. 1, Issue 1, pp. 3–14). https://doi.org/10.1007/s11409-006-6893-0
Vrugt, A., & Oort, F. J. (2008). Metacognition, achievement goals, study strategies and academic achievement: pathways to achievement. Metacognition and Learning, 3(2), 123–146. https://doi.org/10.1007/s11409-008-9022-4
Young, A., & Fry, J. (2008). Metacognitive awareness and academic achievement in college students.
Journal of the Scholarship of Teaching and Learning, 8(2), 1–10.
Zhang, Y., Brady, M., & Smith, S. (2001). Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Transactions on Medical
