When people are asked to evaluate their own performance in a test, their judgments are often not in line with their performance (Dunning et al., 2003, 2004; Couchman et al., 2016; De Bruin et al., 2017; Sanchez and Dunning, 2018; Coutinho et al., 2020). They tend to overestimate themselves and those with the lowest test scores tend to show greater levels of miscalibration (e.g., Kruger and Dunning, 1999; Ehrlinger et al., 2008; Moore and Healy, 2008; Pennycook et al., 2017; Coutinho et al., 2021). Because miscalibration is more pronounced among those who perform poorly in a test, researchers have referred to this finding as the unskilled-unaware effect (Kruger and Dunning, 1999; Dunning et al., 2003). However, whether higher levels of miscalibration indicate lack of awareness about what one knows or does not know is debatable.

Studies looking at overestimation among participants at different levels of performance tend to use a similar method to analyze the data. Researchers first separate the data into quartiles based on participants’ actual test performance, and then compare actual performance to participants’ estimated performance per quartile. This analysis yields a consistent pattern. Overestimation and miscalibration are greater among students in the bottom quartile, and decrease as performance increases. But when performance is very high, a shift to underestimation is observed. Because this finding was first reported by Kruger and Dunning (1999), it is also known as the Dunning-Kruger (DK) effect. The DK effect has since been shown by numerous studies with various tasks including reasoning tests (Pennycook et al., 2017; Coutinho et al., 2021), knowledge-based tests like geography (Ehrlinger and Dunning, 2003), skill-based tests of driving (Marottoli and Richardson, 1998), card gaming (Simons, 2013), sport coaching (Sullivan et al., 2018), and ability-based tests of emotional intelligence (Sheldon et al., 2014).

The primary interpretation of these findings is that low performers overestimate because they are unaware about how poorly they performed due to their own lack of knowledge. Kruger and Dunning (1999) posited that when people have limited knowledge or skills in a domain they are tested in, the lack of knowledge carries a double burden because it prevents people from performing well in the task and from knowing how poorly they did. They proposed that the same knowledge and skills required to do well in a task are also needed for knowing how well one did.

Although the dominant account of these findings is compelling, this is not the only one. Researchers have proposed that this finding could be explained by the hard-easy effect (Juslin et al., 2000), regression to the mean (Feld et al., 2017), regression better than average (BTA) approach (Krueger and Mueller, 2002; Burson et al., 2006), and boundary restrictions (Magnus and Peresetsky, 2022). The hard-easy effect is a tendency that people have of overestimating themselves when a task is difficult and underestimating themselves in easy tasks. In the DK effect, although objective task difficulty does not vary among participants, subjective difficulty varies. The test is difficult for those with low scores but easy for those with high scores. So, according to this account, subjective test difficulty is what drives overestimation. Researchers have also proposed that BTA- the tendency for people to perceive their skills better than average combined with regression to the mean explain the DK effect. Lastly, some researchers have stated that the DK effect occurs because the data is bounded. Based on this view, students in the bottom quartile overestimate because their average performance is relatively low, and performance is bounded at 0. This means that there is great room for overestimation but little for underestimation. For example, if average performance is 30 out of 100, participants can underestimate by 30 points, but they can overestimate by 70 points. So, the likelihood that they will overestimate is higher. The opposite pattern happens for high performers who have little room for overestimation since the maximum score is 100, and therefore are more likely to underestimate themselves.

The DK effect is an observable phenomenon demonstrated by numerous studies. But there are conflicting explanations of it that are supported by empirical evidence. These conflicting accounts are partly due to how calibration accuracy is measured and then compared across performance quartiles. This method is subjected to statistical artifacts like boundary restrictions that are difficult to avoid. Therefore, we argue that using self-estimates and calibration accuracy as indicators of awareness is not ideal. An alternative and more conservative option would be to ask participants to judge the accuracy of their own estimates of performance. That is, ask them to make second-order judgments, which are metacognitive judgments of metacognitive judgments. By doing so and comparing their SOJs with calibration accuracy (estimates of performance minus actual performance), we can get an index of whether they have some insight that their self-estimates are biased. This would be a more conservative indicator of awareness than calibration accuracy per se. According to Kruger and Dunning’s (1999) account, low performers are expected to be quite confident that their self-estimates are accurate because they do not have the means to properly evaluate themselves.

To date a few studies have utilized SOJs to evaluate awareness among students at different performance levels (Miller and Geraci, 2011; Händel and Fritzsche, 2016; Fritzsche et al., 2018; Nederhand et al., 2020). For instance, Miller and Geraci (2011) asked participants to predict their performance in an exam and then rate how confident they were that their self-estimates were well calibrated using a 5-point Likert scale. The results showed that low performing students overestimated how much they got in the test, however, they were less confident that their self-estimates were close to their actual test score than were high performing students. Similar results were found by Händel and Fritzsche (2016) using local metacognitive judgments (judgments for single questions) and postdictions (estimates of performance after completing a test) followed by SOJs. In both studies, low performing students overestimated their performance but their SOJs were lower than high performing participants. These findings thus indicate that low performing students are not as confident about the accuracy of their judgments as high performers, but this does not suggest that they have some level of awareness that their judgments are not well calibrated. Fritzsche et al. (2018) provided evidence for that by reanalyzing the data of Händel and Fritzsche and demonstrating that low performers’ SOJs did not change significantly as a function of how well calibrated their self-estimates were. To know whether low performers have some level of awareness, we need to look at the relationship between SOJs and miscalibration of self-estimates. Nederhand et al. (2020) did exactly that with a sample of college students and another of high school. But in contrast to previous studies, low performers did not show greater levels of miscalibration in terms of overconfidence. In fact, low performers were more accurate in their self-assessments than high performers. Additionally, they found a negative correlation between SOJs and miscalibration in the sample of university students. The more miscalibrated participants’ self-estimates were, the less confidence they had on them. This relationship suggests that participants in the sample were somewhat aware about the accuracy of their self-estimates. A question that arises from this finding is whether such a relationship between miscalibration and SOJs would differ between low and high performers. Based on Kruger and Dunning’s (1999) account of the DK effect, low performers’ SOJs would not increase with a decrease in miscalibration.

The purpose of the present study is to investigate the extent to which students at different levels of performance know that their self-estimates are biased. To do so, university students after completing a regular course exam estimated their performance in the exam, and then rated how confident they were that their self-estimates of performance were accurate by making a SOJ. We then looked at the relationship between SOJs and miscalibration (| estimated performance – actual performance|) among low and high performers separately. It was hypothesized that high levels of miscalibration would be associated with lower SOJs for students who performed well in the exam replicating Nederhand et al.’s (2020) findings. Regarding low performers, if indeed they are not aware of how well they did in the test as proposed by Kruger and Dunning’s (1999), then it is expected that the relationship between miscalibration and SOJs would be positive or nonexistent.

In the present study, we investigated the extent to which low performers know that their self-estimates are inflated. To do so, we asked participants at different levels of performance to rate how confident they were that their self-estimates of performance were accurate. That is, they were asked to make SOJs. We then examined the relationship between miscalibration and SOJs for each quartile separately. It was expected that participants who are aware or somewhat aware about how well they did in the test would show greater confidence in their self-estimates when their self-estimates were closer to their actual test scores. This is exactly what we found for high performers. The less miscalibrated their self-estimates were, the more confident they were in them. However, the opposite pattern was found for low performers. Their SOJs increased with an increase in miscalibration.

The results of this study are in line with Kruger and Dunning’s (1999) interpretation of DK effect. They propose that when individuals are tested on a topic that they have limited knowledge in, they will have difficulty judging how well they did in the test because their ability to accurately evaluate what they know or do not know depends on the same knowledge and skills that led to poor performance. The current findings complement this account by showing that failings in metacognition are not easily detectable by the learner. But why is that the case? One possible explanation for this finding is that when individuals have limited knowledge about a topic they are tested on, they are less likely to have access to valid metacognitive cues—that is, cues that are predictive of performance. Therefore, they end up relying on surface-based (or fluency-based) cues such as retrieval fluency (the ease with which information comes to mind) or processing fluency. The problem is that cues like fluency predict performance only sometimes, mainly when changes in fluency coincide with variations in valid cues. For example: in a memory test, when increases in retrieval fluency is associated with increases in the strength of the memory signal (valid cue), retrieval fluency accurately predicts performance. Otherwise, it does not. In line with this, previous research has shown differences in cue utilization among students at different levels of performance (Thiede et al., 2010; Ackerman and Leiser, 2014; Gutierrez de Blume et al., 2017). For instance, Ackerman and Leiser (2014) demonstrated that low achievers – when regulating their learning of text—were more sensitive to unreliable, surface-level cues (e.g., ease of processing, readability, and specific vocabulary) than high achievers. Additionally, Thiede et al. (2010) showed that students who were at-risk readers were more likely to base their judgments of text-comprehension on surface-based cues. Conversely, competent readers relied on valid comprehension-based cues, such as the belief they can explain the text to another person. The results of these studies taken together suggest that when participants are less knowledgeable about the material they are tested on, they are more likely to rely on fluency-based cues and believe that such cues are good predictors of performance when in fact they are not.

We also found that estimated scores were a stronger predictor of SOJs than actual test scores for every quartile. This suggests that SOJs may depend on similar mechanisms as self-estimates. But it is likely that the cues that low performers rely on when making metacognitive and second-order judgments are not the same as these utilized by high performers and this is why for low performers confidence increases with miscalibration whereas for high performers confidence decreases with miscalibration.

Additionally, we compared SOJs between lower and higher performers, and we found differences between the two groups. Lower performers displayed less confidence in the accuracy of their estimates than more proficient peers. These findings are consistent with previous studies showing low performers’ general tendency to give lower confidence ratings (see Fritzsche et al., 2018). This finding could be also attributed to differences in cue utilization between low and high performing students. Although low performers rely on surface-based cues and are confident on them, such cues are unlikely to give rise to the same level of confidence as valid cues because they are unreliable.

We provided evidence that an increase in miscalibration is associated with an increase in SOJs for low performers. This supports and complements the dual burden account of the DK effect. We suspect that this may be the result of how low performers interpret cues available during testing. Specifically, relying on cues that are fluency-based and not reflective of performance. It is likely that low performing students view increases in fluency as an indicator that their responses are correct, which then lead to inflated self-estimates and higher confidence in their self-estimates. The present study has theoretical implications to the literature on the DK effect, and practical implications to education. First, it shows that low performers lack awareness that their estimates are biased. Second, it suggests that proper calibration among low performers is unlikely to happen by itself. Hence, it is important that in school, students learn about these biases, its effects in learning and performance and what they can do to minimize its influences.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: OSFHOME repository: https://osf.io/nqrhy/files/osfstorage/64a313b2a2a2f4128543694c.

The study involving human participants was reviewed and approved by Zayed University, UAE. The participants provided their written informed consent to participate in this study. Ethical clearance was obtained from Zayed University Research Ethics Committee (ZU19_103_F).

MC contributed to conceptualization, review of the literature, design and methodology of the study, data analysis and interpretation, drafting of the manuscript, submission and review process. JT contributed to the literature review, data collection, and revision of the draft. IF-L contributed to recruiting participants, data collection, and data entry. SA contributed to the writing of the literature review and methods. JC contributed to drafting of the manuscript. All authors contributed to the article and approved the submitted version.