Influence of Domain Knowledge on Monitoring Performance Across the Life Span

Two studies were conducted to investigate effects of domain knowledge on metacognitive monitoring across the life span in materials of different complexity. Participants from 4 age groups (3rd-grade children, adolescents, younger and older adults) were compared using an expert–novice paradigm. In Study 1, soccer experts' and novices' ease-of-learning judgments (EOLs), judgments of learning (JOLs), and confidence judgments (CJs) were contrasted when memorizing soccer-related word pairs. In Study 2, monitoring judgments (i.e., a rating of global comprehension, JOLs, and CJs) were collected in regards to a soccer-related narrative. The results of both approaches showed that experts' better memory performance obtained in both studies was not always accompanied by advantages in monitoring performance. In Study 1, experts of all ages outperformed novices in monitoring accuracy. In Study 2, no benefits of expertise on monitoring were found; in children, novices even surpassed experts in monitoring quality. In both studies, the most consistent influence of previous domain knowledge on monitoring performance concerned more optimistic judgments of experts compared with novices, regardless of stimuli and recall format. In sum, our results document a twofold effect of expertise on monitoring. Although domain-specific knowledge enhances monitoring performance in some situations, more optimistic estimates, presumably due to the application of a familiarity heuristic, typically reduce experts' monitoring accuracy.

Evidence for the powerful effects of domain knowledge on memory performance comes from studies using the expert–novice paradigm. These studies compared experts and novices in a given domain (e.g., baseball, chess, or soccer) on a memory task related to that domain. Perhaps the most robust finding in the literature on knowledge effects is that experts in an area learn faster and more when studying information in their domain of expertise than do novices (e.g., Feltovich, Prietula, & Ericsson, [15]; Simon & Chase, [48]). From a developmental perspective, the major advantage of the expert–novice paradigm is that knowledge and chronological age are not necessarily confounded, a problem inherent in most studies addressing knowledge-base effects. Several studies have demonstrated that rich domain knowledge enabled a child expert to perform much like an adult expert and better than an adult novice—thus showing a disappearance and sometimes a reversal of usual developmental trends (e.g., Nietfeld & Schraw, [37]; Schneider, Gruber, Gold, & Opwis, [43]). It has also been shown that previous knowledge also reduces deficits related to aging. For example, professional expertise may even overcome the associative deficit usually found in older adults (Castel, [6]; Naveh-Benjamin, [34]). Despite these impressive cognitive benefits of even quite young experts within the scope of cognitive research, their advantage regarding metacognitive monitoring performance is less clear. Although several developmental studies have shown that experts in a specific domain possess more declarative, verbalizable metacognitive knowledge in this domain (e.g., Schneider, Körkel, & Weinert, [44]; Schneider, Schlagmüller, & Visé, [45]), the situation regarding monitoring skills bears conflicting evidence and therefore is addressed in the following two studies.

According to Nelson and Narens's (1990) seminal model of procedural metamemory, monitoring encompasses several prospective and retrospective judgments at different stages of the learning process. Of special interest for the following two studies are ease-of-learning judgments (EOLs) as indicators of self-perceived task difficulty, which are collected before the actual learning phase begins. Second, judgments of learning (JOLs), which take place during or shortly after learning, were included. They represent subjective ratings about the degree to which encoded information can be potentially recalled during a future memory test (Nelson & Narens, [36]). Finally, confidence judgments (CJs) were collected: They concern retrieval monitoring and are typically made after a response is given to indicate how sure participants are about the correctness of an answer. CJs are thought to reflect a substantive sense of certainty that arises from the strength of the memory that is being retrieved, and this sense of certainty has been interpreted as an indicator of memory accuracy (Ghetti, Lyons, Lazzarin, & Cornoldi, [17]; Roebers, [40]). As very few studies exist that directly compare different monitoring measures, our approach is designed to provide more a profound insight into how previous knowledge affects monitoring at different stages of the learning process.

There are two fundamentally different ways to assess and analyze the quality of monitoring processes. On the one hand, absolute accuracy (calibration) refers to the extent to which the mean monitoring judgment represents the mean level of performance, thus indicating whether overconfidence versus underconfidence is predominant (cf. Finn & Tauber, [16]; Schwartz & Efklides, [46]). So far, calibration approaches repeatedly revealed for different age groups that learners tend to be more optimistic regarding their own performance regardless of age (e.g., Dunlosky & Rawson, [12]; Koriat, [26]). Yet, the empirical basis is still scarce. Schneider et al. ([44]), who compared soccer experts and novices in third, fifth, and seventh grade, found that experts outperformed novices in metacognitive performance (JOLs and feeling-of-knowing judgments), thereby resulting in less overconfidence.

On the other hand, relative accuracy (resolution) refers to the extent to which participants are able to allocate higher monitoring judgments to items that are more likely to be correctly remembered and lower judgments to items that are less likely to be remembered (e.g., Schwartz & Efklides, [46]). Relative accuracy can be determined by contrasting the mean metacognitive judgments in correct versus incorrect answers (differentiation performance; e.g. Von Der Linden, Schneider, & Roebers, [56])—an approach that allows for tracing back the absolute height of the judgments. A more common procedure used in several studies is to calculate Goodman-Kruskal gamma correlations between judgments and recall performance. A positive correlation indicates that higher judgments were given for items that were recalled correctly than for those recalled incorrectly (accuracy; cf. Koriat, Ackerman, Adiv, Lockl, & Schneider, 2014; Nelson, [35]).

The majority of studies examining the effect of a rich knowledge base on metacognitive monitoring so far have investigated differences in relative accuracy between experts and novices and mainly, but not exclusively, revealed advantages in favor of experts. For instance, de Bruin, Rikers, and Schmidt ([8]) found that chess experts not only learned endgame moves faster than less experienced players but also gave more accurate JOLs when predicting their performance in the following move. In the baseball domain, Griffin, Jee, and Wiley ([19]) revealed advantages in favor of participants with more domain knowledge in terms of absolute but not of relative JOL accuracy when studying baseball texts. For CJs, Nietfeld and Schraw ([37]) showed that younger adults with more experience in statistics monitored their confidence more accurately when solving probability tasks than did their novice peers. However, a classical study applying the resolution approach also has revealed an inverse relationship between expertise and metacognitive monitoring when judging the comprehension of texts. That is, the richer the domain knowledge (physics vs. music), the less accurate the CJs were in this student sample (Glenberg & Epstein, [18]). In one of the few studies including other age groups, Toth, Daniels, and Solinger ([52]) compared younger and older adults' metacognitive monitoring using stimuli of different levels of familiarity—an approach that can roughly be related to the expert–novice paradigm because previous knowledge at baseline served as the dependent variable. They asked younger and older adults to rehearse names of actors either famous in the 1950s or in the 1990s. Their findings showed that—despite the expected benefit on memory performance for the respective era—older adults' JOL accuracy was not enhanced by previous knowledge: Their tendency to give rather high JOLs when rating later memory performance for actors from the 1950s resulted in lower JOL accuracy.

These advantages of experts can be explained by the fact that they can activate a greater knowledge base when giving their monitoring judgments than can novices (Griffin et al., [19]). In addition, experts should process information in a more automatized manner, and tasks in the specific domain should be less demanding for them. Thus, experts should possess more cognitive resources to monitor their performance (de Bruin et al., [8]; Nietfeld & Schraw, [37]). In addition, it is argued that novices commit more errors, which impairs their overall monitoring accuracy as they have to put more effort into task monitoring (de Bruin et al., [8]). On the other hand, one possible reason for experts' more optimistic estimates of performance compared with novices may be that participants use a self-classification heuristic and thus rely less on information arising from the presented material (Glenberg & Epstein, [18]). More fine-grained explanations arise from the cue-utilization approach (Koriat, [25]): Especially experts may base their monitoring judgments on cues as ease of processing, fluency at retrieval, or cue familiarity in general (Finn & Tauber, [16]; Kelley & Sahakyan, [24]; Metcalfe & Finn, [33]) might lead to more optimistic estimates and consequently be less valid for later performance.

In sum, the results reported here leave many questions unanswered. First, as most of these studies only included one single monitoring measure (for an exception, see Schneider et al., [44]), it is not possible to directly compare the effects of expertise on different measures. Yet, partly different underlying cues have been postulated for EOLs, JOLs, and CJs suggesting that expertise might have differential effects on the respective judgments as well. As EOLs are collected before the learning phase, they are not only affected by perceived complexity and concreteness (cf. Jönsson & Lindström, [23]), but they are also based on prior experiences with the particular domain (Mazzoni, Cornoldi, Tomat, & Vecchi, [32]). Thus, more optimistic estimates of experts compared to novices should be most pronounced in EOLs. In contrast, JOLs have been both linked to item characteristics such as familiarity (Nelson & Narens, [36]) and perceived difficulty (Koriat & Bjork, [28]) or—in the case of delayed JOLs—to characteristics of retrieval such as fluency (Undorf & Erdfelder, [53]). Consequently, experts' optimism should be lower for JOLs than for EOLs, and JOL accuracy should be enhanced as compared with those novices. Finally, CJs are found to depend primarily on retrieval characteristics—for example, on perceived ease (Zakay & Tuvia, [57]) or vividness of retrieval (Robinson, Johnson, & Robertson, [39]). Thus, greater accessibility of items in experts' networks might even enhance their accuracy in contrast to novices (cf. also Koriat, [26]). So far, no approach compared the impact of expertise on several monitoring measures within one analysis. Thus, our study is designed to fill this gap.

Another question that has arisen from recent results concerns the issue of whether expertise affects monitoring performance differentially in different stages of life. The inclusion of a life-span sample is of particular interest because even in elementary schoolchildren, previous knowledge enhances memory performance in both quantitative and qualitative aspects (e.g., Schneider et al., [43]). Still, increases in domain knowledge are considerable from childhood to adolescence (Schneider, [42]). In older adults, previous knowledge has been shown to compensate for age-related differences in memory performance compared with younger adults (e.g., Hultsch & Dixon, [22]; Toth et al., [52]). The discussed mechanisms underlying this advantage of participants with greater domain-related knowledge seem to be similar in different age groups (greater semantic network, more efficient strategic processing, chunking; cf. Schneider, [42]). It has to be noted, however, that in children and older adults, these benefits occurred only in situations where no misleading questions or contradictory information was involved (Castel, [5]; Elischberger, [14]). In sum, the impact of previous knowledge on memory performance across the life span seems to follow a complementary developmental trajectory. The question remains unsolved as to whether these findings can be directly transferred to monitoring processes as comparisons of broader age ranges are still missing. This extension is considered important because it has been shown several times for participants from primary school to older adulthood that accurate monitoring correlates with memory performance as well (e.g., Dunlosky, Kubat-Silman, & Hertzog, [10]; Schneider, [41]; Thiede, [50]). However, developmental studies examining the link between previous knowledge and metacognitive monitoring are still scarce. Results for primary school children suggest that even third graders benefit from previous knowledge in terms of metacognitive monitoring (Schneider et al., [44]). Thus, an even broader knowledge base in adolescence and somewhat enhanced general monitoring ability in this age group (cf. Hoffmann-Biencourt, Lockl, Schneider, Ackerman, & Koriat, [21]) might have further favorable effects in experts. In younger adults, the majority of the results indicate that experts surpass novices in monitoring performance (for an exception, see Glenberg & Epstein, [18]). In this age group, in most cases, experts have more cognitive resources for the monitoring process; they are able to distinguish between newly acquired information and already existent memory content and thus use diagnostic cues for their predictions and postdictions. In older adults, one might assume that these positive outcomes for experts cannot fully be maintained because they are found to rely too heavily on item familiarity and previous experiences within their domain of expertise and therefore are too optimistic about their own performance (cf. Shing, Werkle-Bergner, Li, & Lindenberger, [47]; Toth et al., [52]). As only a few studies so far have dealt with possible effects of expertise on monitoring performance in different stages of life, the cited results only indicate possible developmental trajectories.

In summary, the question of under which circumstances a broader knowledge base represents an advantage or a disadvantage when it comes to metacognitive monitoring still remains unanswered. Various factors such as the age group under study, the operationalization of monitoring, and the time of measurement during the learning and retrieval process may explain differences in the existing literature.

In the research to be presented, we aimed to explore the importance of these factors—that is, age group and kind of metacognitive monitoring—on experts' and novices' memory performance and procedural metamemory. We also made sure that the memory tasks were neither too difficult for novices nor too easy for experts to avoid floor and ceiling effects. Experts and novices in four age groups (children in third grade, adolescents in seventh and eighth grade, younger and older adults) were included in our experiments because we assumed that expertise operates differently on monitoring in different stages of cognitive development. We opted for third graders as the youngest age group because at this age, children are supposed to have acquired a reasonable amount of knowledge in their domain of expertise (Schneider et al., [44]) and because, in general, their monitoring judgments are already relatively accurate (Schneider, [42]). Monitoring performance improves only slightly from middle childhood to adolescence (Hoffmann-Biencourt et al., [21]). On the other hand, knowledge in the domain of soccer typically undergoes substantial increases in the same period (Schneider et al., [44]). Thus, we decided to include third graders as well as seventh and eighth graders in the sample to represent the age range from childhood to adolescence. Given the lack of studies that have compared performance of younger and older adults on such tasks, these two age groups were also included in our design. There is research indicating a decline in strategic and associative components of episodic memory in the latter age group, which may lead to difficulties in learning new information comparable to those demonstrated by 10-year-old children (Shing et al., [47]). At the same time, monitoring performance has been found to remain rather intact in older adults (Dunlosky & Metcalfe, [11]). One interesting issue discussed in the relevant literature concerns possibly diverging effects of expertise on monitoring in children and older adults. Although domain-specific knowledge can delay memory impairments in older adults (Castel, [6]), expertise seems to lead to illusions of knowing rather than to improved monitoring performance in this age group, which is not true for children (Toth et al., [52]). To directly compare differential effects of a rich knowledge base on monitoring, participants of a broad age range were included. To our knowledge, no study has examined the effect of prior knowledge on metacognitive monitoring across the life span. Soccer was chosen as the domain because we expected to be able to find soccer experts and novices in every age group from elementary school onward.

In Study 1, the focus was on basic monitoring processes. Learning materials consisted of soccer-related pairs of concrete nouns. Monitoring performance was analyzed by means of the resolution approach to enhance comparability with previous studies. Gamma correlations were computed as an indicator of accuracy; in addition, we analyzed differentiation performance—that is, the extent to which mean JOLs differed between correct and incorrect responses—to reflect the absolute height of the judgments and to obtain an indicator of the degree of optimism in judgments, which is not possible with gamma correlations.

A total of 160 participants (93 male, 67 female) in four age groups (40 children in third grade, 40 adolescents in seventh and eighth grade, 40 younger adults, and 40 older adults) took part in our study. Participants came predominantly from a middle-class socioeconomic background. They were recruited via direct school contact and via newspaper and Internet advertisements. Participants received small gifts or vouchers or were paid cash. As expected, gender was not equally distributed across experts and novices: There were 66 male and 14 female experts. To date, the vast literature on metacognitive processes does not discuss any effects of gender, neither in different age groups nor in different aspects of metacognition (Dunlosky & Metcalfe, [11]; Rhodes & Tauber, [38]; Schneider, [42]). Moreover, the few studies that have explored differences in expertise in male and female soccer experts did not find any differences (e.g., Schneider et al., [44]). To validate this finding, we explored the effect of gender in several preliminary analyses for experts and novices, using gender as a between-subjects factor, and did not find systematic gender differences. Thus, we collapsed data across this variable. Mean age was 8;4 (SD = 6.36 months), 12;11 (SD = 8.88 months), 23;6 (SD = 40.56 months), and 66;10 (SD = 64.56 months) for children, adolescents, younger adults, and older adults, respectively.

The learning items consisted of 22 pairs of soccer-related concrete nouns, half of which were semantically associated (e.g., "field—stadium") and half not associated (e.g., "winner—pass"). Results of a pilot study with 43 participants served as a basis for the selection of the item list. The order of presentation was randomized among the participants.

We further applied a test assessing soccer knowledge based on an instrument developed by Schneider et al. ([44]). The items concerned both knowledge of soccer rules and important soccer events. The knowledge tests differed for children and adults. Children were asked 9 questions that were easier than those asked of the other age groups to avoid floor effects. The other age groups responded to 11 questions.

Consent was obtained from parents and the school for minor participants before the beginning of the study. Participants were tested individually by trained experimenters in quiet rooms in the school or in the laboratory.

The word pairs were presented on a computer screen one at a time. For the EOL rating, participants had to assess in a self-paced way for each pair how easy or difficult they would find it on a scale ranging from 1 ("very difficult") to 7 ("very easy"). The scale was marked with a sad face on the left-hand side, a neutral face in the middle, and a happy smiley face on the right-hand side. After that, items were presented again in the same order to study. Presentation rates were 8 s per item pair for children and older adults, 6 s for adolescents, and 3 s for younger adults. The presentation rates were obtained on the basis of the pilot study to control for baseline difficulty between the age groups. Participants were instructed to study the pairs and were told that they would have to remember the target when presented with the stimulus. In the JOL phase, each left noun of the item pair was presented on the screen in the same order as in the learning phase. Participants were asked to indicate the likelihood of remembering the target in about 30 min. JOLs were rated on a similar scale as in the EOL phase marked with smiley faces, which ranged from 1 ("very unlikely") to 7 ("very likely"). After the JOL phase, the soccer questionnaire and the demographic questionnaire for younger and older adults were administered.

In the recall phase, the left word of each pair was presented. Participants had to indicate their answer and were asked to guess if they did not know the word. After each response, they rated how confident they were about their answer. CJs were measured accordingly to EOLs and JOLs on a 7-point smiley face scale from 1 ("very unsure") to 7 ("very sure").

The main measures of interest consist of recall rate, EOLs, JOLs, and CJs. We first present memory performance as the percentage of correctly recalled items. Then we will analyze monitoring performance. Results are reported as a function of age group, monitoring measure, and expertise to examine the effect of previous knowledge on metacognitive monitoring in different stages of life. To check the assumptions for using an analysis of variance (ANOVA), the variables were tested for normal distribution by means of the Kolmogorov-Smirnov test and for homogeneity by means of Levene's test. Normal distribution was given for each variable. Homogeneity was given for the differentiation analysis; in gamma correlations, this assumption was only fulfilled in EOLs. To allow for comparability of the applied tests, the ANOVA was also used to analyze monitoring accuracy as this method is rather robust provided that sample sizes are equal and large enough (Bortz, [3]; Bühner & Ziegler, [4]). Both requirements are met in the following analysis. As a post-hoc follow-up on main effects, Scheffé tests were used. Level of significance was set to p < .05. To estimate the magnitude of significant results, partial eta square values are reported as effect sizes.

Results in the soccer questionnaire were used to assign participants to the expert versus novice condition. Participants with percentages of correct answers above the median in their age group were classified as experts, and those with percentages of correct answers below the median were classified as novices. T tests revealed that the difference in soccer knowledge between novices and experts was highly significant in each age group (all ps <.001). In detail, the percentages of correct answers in the soccer test were 22.22% (SD = 18.73%) and 65.56% (SD = 16.48%) in children, 15.50% (SD = 10.99%) and 51.50% (SD = 18.72%) in adolescents, 25.50% (SD = 13.95%) and 79.00% (SD = 15.53%) in younger adults, and 30.00% (SD = 20.52%) and 77.50% (SD = 8.51%) in older adults.

Mean recall rates in percent as a function of age group and expertise are depicted in Table 1. An ANOVA with expertise and age group as between-subject factors revealed a main effect of expertise, F(1, 152) = 25.45, p < .001, ηp2 = .14, with experts (.51) recalling more items correctly than novices (.38), and a main effect of age group, F(3, 152) = 8.32, p < .001, ηp2 = .14, with post-hoc tests showing that children (.35) and older adults (.40) recalled fewer items correctly than adolescents (.51) and younger adults (.50). Furthermore, an interaction between expertise and age group was found, F(3, 152) = 3.35, p < .05, ηp2 = .06. Separate ANOVAs for each age group revealed that it was only in older adults that experts and novices did not differ in recall rates.

Table 1 Recall rates as a function of age group and expertise

5 Note. Standard deviations and ranges are in parentheses.

To assess monitoring performance, we first analyzed to what extent participants were able to differentiate between correct and incorrect responses. This is typically done by calculating mean ratings before and after correct versus incorrect responses (e.g., Destan, Hembacher, Ghetti, & Roebers, [9]; Von Der Linden et al., [56]; cf. Table 2). An ANOVA with age group and expertise as between-subjects factors and correctness of response and monitoring measures as within-subjects factors revealed a significant main effect of age group, F(3, 152) = 5.97, p < .01, ηp2 = .11. Post-hoc tests showed that children gave higher mean ratings (4.84) than adolescents (4.34) and younger adults (4.37). Furthermore, a significant main effect of expertise was revealed, F(1, 152) = 23.98, p < .001, ηp2 = .14, indicating that experts gave higher mean ratings (4.74) than novices (4.28). Finally, the main effects of correctness of response, F(1, 152) = 1,189.46, p < .001, ηp2 = .89, and monitoring measure, F(2, 304) = 30.81, p < .001, ηp2 = .17, as well as the interaction between correctness of response and monitoring measure, F(2, 304) = 383.66, p < .001, ηp2 = .72, and between age group and monitoring measure, F(6, 304) = 2.89, p < .05, ηp2 = .05, turned out significant. For the interaction between correctness and monitoring measure, paired t tests comparing mean EOLs, JOLs, and CJs in correct answers revealed that mean ratings were lower in EOLs (5.27) than in JOLs (5.55; p < .05) and lower in JOLs than in CJs (5.95; p < .001). In incorrect answers, we found the reverse pattern: Mean ratings were higher in EOLs (4.53) than in JOLs (3.76; p < .001) and higher in JOLs than in CJs (2.50; p < .001). That is, in all monitoring measures, participants were able to differentiate between correct (5.59) and incorrect responses (3.43); however, in CJs, this difference was largest, followed by JOLs and EOLs. To interpret the interaction between age group and monitoring measure, we calculated separate ANOVAs with monitoring measure as a within-subjects factor. In adolescents and younger adults, we did not find any significant effect of monitoring measure. In children, the main effect of monitoring measure turned out significant, F(2, 78) = 30.10, p < .001, ηp2 = .44. Paired t tests revealed that children's mean EOLs (5.44) were higher than their JOLs (4.30; p < .001) and their CJs (4.01; p < .001). In older adults, about the same pattern resulted: The ANOVA showed a significant main effect of monitoring measure, F(2, 78) = 15.92, p < .001, ηp2 = .29. Paired t tests indicated that older adults gave higher mean EOLs (4.86) than JOLs (4.04; p < .01) and CJs (3.79; p < .001) and higher mean JOLs than CJs (p < .05). No further interactions reached the significance level.

Table 2 Mean EOLs, JOLs, and CJs in correct versus incorrect responses as a function of age group and expertise (Study 1)

6 Note. EOLs = ease-of-learning judgments; JOLs = judgments of learning; CJs = confidence judgments. Standard deviations are in parentheses.

To assess monitoring accuracy as a function of age group, expertise, and monitoring measure, Goodman-Kruskal gamma correlations between each monitoring judgment and recall performance were computed for each participant and then averaged for each single cell in the experimental design. Gamma correlations are considered to be the most appropriate measure of metacognitive accuracy (Nelson, [35]) and are commonly used in contemporary literature (Koriat, Ma'ayan, & Nussinson, [30]; Von Der Linden & Roebers, [55]). Two-tailed t tests showed that all correlations displayed in Table 3 are different from 0 (EOLs, ps <.05 for children, ps <.01 for adolescents and novices among older adults, ps <.001 for younger adults and experts among older adults; JOLs and CJs, all ps <.001).

Table 3 Mean gamma correlations as a function of age group and expertise

7 Note. EOLs = ease-of-learning judgments; JOLs = judgments of learning; CJs = confidence judgments. Standard deviations are in parentheses.

The ANOVA with age group and expertise as between-subjects factors and monitoring measure as a within-subjects factor revealed a significant main effect of expertise, F(1, 148) = 5.32, p < .05, ηp2 = .04, indicating that experts (.71) had higher mean correlations than novices (.64). Furthermore, significant main effects of age group, F(3, 148) = 4.30, p < .01, ηp2 = .08, and of monitoring measure, F(2, 296) = 146.23, p < .001, ηp2 = .50, were found and were qualified by a significant interaction between both, F(6, 296) = 3.23, p < .01, ηp2 = .06. Separate ANOVAs for each monitoring measure with age group as a between-subjects factor showed that only in EOLs, age group had a significant effect on the height of the gamma correlations, F(3, 152) = 4.55, p < .01, ηp2 = .08, in that children (.25) had lower mean correlations than younger adults (.56). In general, paired t tests revealed that gamma correlations were lowest for EOLs (.39) followed by JOLs (.75) and CJs (.90; all ps <.001). The remaining interactions missed the significance level.

To emphasize the validity of the results, we repeated the analyses with a stricter classification of expertise: We removed participants with soccer knowledge scores in the middle third. Thus, participants who were better than two thirds of the sample in soccer knowledge were compared to participants who did worse than two thirds of the sample. Given that results were completely in accord with the findings reported already, we do not present them here in greater detail.

The first experiment aimed to shed light on the controversial debate about whether soccer experts benefit from their previous knowledge in terms of monitoring quality in a paired-associates paradigm. The existing literature includes findings that either indicate experts' advantages concerning monitoring performance (e.g., Griffin et al., [19]; Nietfeld & Schraw, [37]) or that state the effect of previous knowledge leads to increased optimism in experts and thus reduces monitoring accuracy (e.g., Glenberg & Epstein, [18]). A second aim was to expand the use of expert–novice paradigms to a life-span approach. There are very few studies investigating monitoring processes in participants with different levels of previous knowledge in children, adolescents, or older adults (cf. Schneider et al., [44]; Toth et al., [52]), and there is—to our knowledge—not a single one comparing findings across a broader age range. A third important objective was to contrast the effects of expertise using different monitoring measures.

First of all, we found that the adjustment of presentation times in the different age groups successfully avoided ceiling and floor effects as mean recall rates ranged from 0.3 to 0.6. In line with the existing literature, participants with a more profound soccer knowledge outperformed participants with less domain knowledge in recall (e.g., Nietfeld & Schraw, [37]; Schneider et al., [44]). However, this finding did not hold true for the oldest age group: On closer inspection, elderly experts surpassed their novice peers in memory for associated items, but not for unrelated items; here, senior experts performed somewhat worse than novices. These descriptive findings can be explained with the "associative deficit hypothesis" (Naveh-Benjamin, [34]), according to which older adults manifest difficulties in creating new links between originally unrelated items. It may be especially true for experts in this age group as they come with more persistent bindings between soccer-related words than do novices. Overall, material and design proved to be appropriate.

As far as metacognitive monitoring is concerned, participants of all age groups were able to anticipate the potential ease of the later learning process, to predict their later performance, and to differentiate between remembered and forgotten items. Participants' differentiation performance and accuracy were best in CJs, followed by JOLs and finally EOLs. This finding is in line with previous findings indicating that CJs in paired-associates paradigms are usually more precise than JOLs (cf. Koriat & Goldsmith, [29]; Rhodes & Tauber, [38])—most likely because they are based on cues of both the learning phase and the retrieval phase. Given that EOLs take place just before the actual encoding process, they tend to be less accurate than later measures (Leonesio & Nelson, [31]). Regarding both EOL differentiation and accuracy, it appears that children and—to a smaller extent—older adults experienced somewhat more difficulties with anticipating the upcoming learning process, which may be due to the fact that they deal less often with abstract stimuli in daily life than the other two age groups. In general, our findings are consistent with studies revealing that JOLs and CJs underlie minor age effects from late elementary school onward (Schneider, [42]) and that declines in quality in older adults are rare (Hertzog & Dunlosky, [20]).

The main focus of Study 1 was the effect of expertise on monitoring quality in different stages of life. Overall, our data point in the direction of work stating that experts tend to be more optimistic about their performance (Glenberg & Epstein, [18]; Son & Kornell, [49]). In the differentiation measures, a general main effect of expertise was found, indicating that experts gave higher monitoring ratings than did novices. Interestingly, this finding was true for each of the four age groups. Thus, we cannot confirm our assumption that more optimistic judgments are more pronounced in experts among older adults. However, for monitoring accuracy, an advantage for experts was revealed, regardless of age. Participants with richer soccer knowledge had higher mean gamma correlations than novices, indicating that experts monitored their own performance more accurately than novices (cf. also Nietfeld & Schraw, [37]). Although the same tendency was also found in differentiation performance, it missed the significance level (p < .1).

In general terms, although the impact of expertise on memory has been confirmed in the present study, the impact of previous knowledge on metacognitive monitoring across the life span is not similarly obvious: Our findings show that the superior outcomes for experts were found for accuracy, but not for differentiation. A possible limitation of our approach is the fact that results from paired-associates designs are not fully transferable to metacomprehension and thus more everyday-life-like settings (e.g., Thiede, Griffin, Wiley, & Redford, [51]). This question is addressed in our second experiment. Here, we also investigated experts' and novices' metamemory performance in a metacomprehension approach dealing with monitoring comprehension of a cohesive soccer-related narrative.

In the second study, participants had to judge their own performance when answering questions on a soccer-related narrative. Again, we emphasized the use of different metacognitive measures to facilitate comparisons with the first experiment and to further investigate whether expertise affects different monitoring indexes differently. In general, we expected to be able to replicate our previous results in that experts should consider their prospective and retrospective performance more positively than novices; an advantage in favor of experts should only be reflected in monitoring accuracy.

Again, a total of 160 (82 male, 78 female) participants of the same four age groups (40 children in third grade, 40 adolescents in seventh and eighth grade, 40 younger adults, and 40 older adults) participated in this experiment. Recruitment procedures, socioeconomic background, and gratifications were the same as in Study 1. Again, gender distribution differed between experts and novices as there were 64 male and 16 female experts. Participants' mean age was 8;2 (SD = 4.9 months), 13;0 (SD = 11.4 months), 21;11 (SD = 33.84 months), and 69;1 (SD = 58.2 months) for children, adolescents, younger adults, and older adults, respectively.

The story and the recall questions were chosen on the basis of extensive previous piloting. The study was based on a story about a boy named Max who attended a soccer match with his father. The text yielded information about several external facts—for instance, how Max and his father got to the stadium or that the match coincided with Max's birthday—and facts that were directly related to the events on the soccer field. Some relevant information concerning the soccer game had to be inferred from the text—for instance, the absolute number of goals scored during the match. Though soccer knowledge was expected to facilitate the understanding of the text, it was not required to draw the correct inferences. The plot loosely followed a narrative by Schneider et al. ([44]). However, story length and complexity were adapted to the four different age groups in that children received the shortest text (626 words). Adolescents and older adults were given a text of medium length (795 words), and younger adults were given the longest story (991 words). Complex details were omitted for children. The text was presented in taped and written format.

Memory performance was collected by means of open questions on the text. The difficulty of the questions was balanced between the different age groups on the basis of previous piloting. We ensured that all participants received simple, medium, and difficult questions limited to 18 in children and 20 in all other age groups. Monitoring was measured by one global judgment of understanding and by predictions of later performance after each question on the text (JOLs) and postdictions after the responses (CJs). Note that we chose the terms "JOLs" and "CJs" in this metacomprehension approach instead of the more common terms "text prediction" or "metacomprehension judgment" (Dunlosky, Rawson, & McDonald, [13]) to boost comparability with Study 1.

The assessment of soccer knowledge and assignment to the experimental conditions were identical to those realized in Study 1.

As in the first experiment, participants were tested individually. For minor participants, consent of parents and of the school was obtained. Participants first listened to the narrative via headphones, and then they listened to the audio version and were asked to read the written text at the same time. Participants were told beforehand to listen carefully because they were going to be asked questions about the soccer story. After that, they were to give a rating on a 7-point smiley face scale to indicate how well they understood the story's meaning, with a sad face and the words "very poorly" on the one end and a happy smiley face labeled "very well" on the other (global judgment). Participants were then requested to fill out a JOL questionnaire with the questions on the text described earlier. Each question started with the phrase, "How sure are you that you are going to remember that ...". Thus, participants noted their JOL prediction by using a similar 7-point scale this time labelled "very unsure" to "very sure." After the JOL phase, the tests of verbal and fluid intelligence and the demographic questionnaire for younger and older adults were administered to their respective groups. In the recall phase, the same questions were asked again. Participants were asked to write their answers and give ratings of subjective confidence after each response (again ranging from 1 to 7).

The main measures of interest are recall rate, global judgment, JOLs, and CJs.

We first present memory performance data in terms of percentage of appropriate responses to questions on the text. Then we analyze the predictive value of the global judgment. Finally, JOLs and CJs as indicators of metacognitive monitoring will be examined. Results are reported as a function of age group and expertise. Statistical analyses were carried out similarly to those in the first experiment. The assumptions for the ANOVA (normal distribution and homogeneity) were tested. Normal distribution was given for each variable except for gamma correlations in JOLs. Homogeneity was given for the differentiation analysis; in gamma correlations, this assumption was met for all variables except for JOLs. To allow for comparability of the applied tests, the ANOVA was also used to analyze monitoring accuracy as this method is rather robust provided that sample sizes are equal and large enough (Bortz, [3]; Bühner & Ziegler, [4]). Both requirements are met in the following analysis. Again, the assignment to the expert condition versus novice condition was carried out according to the median in soccer knowledge. Experts and novices differed significantly in mean percentages of correct answers in the soccer tests (ps <.001 in t tests for each age group). In children, means resulted in 13.89% (SD = 8.74%) and 61.67% (SD = 24.84%). Adolescents attained 15.91% (SD = 9.73%) and 65.00% (SD = 25.42%); younger adults attained 36.82% (SD = 11.97%) and 86.36% (SD = 13.35%); and older adults attained 16.82% (SD = 11.90%) and 71.36% (SD = 19.18%). Results were collapsed across gender as preliminary analyses for experts and novices did not reveal any effects of gender within each experimental group.

In Table 1, the percentages of correct responses are depicted in the second section. An ANOVA with expertise and age group as between-subjects factors revealed a main effect of expertise, F(1, 152) = 25.97, p < .001, ηp2 = .15, with experts (.59) giving more correct responses than novices (.46). Furthermore, there was a significant main effect of age group, F(3, 152) = 8.46, p < .001, ηp2 = .14. Post-hoc tests showed that adolescents (.49) and older adults (.45) had fewer correct responses than younger adults (.61). Children's performance did not differ significantly from that of the three other age groups. The interaction between both factors was not significant.

To assess the accuracy of the global judgment, bivariate correlations between the global judgment and later recall performance were calculated. The results indicate that correlations were generally very close to 0 with coefficients in experts being.016 for children,.048 for adolescents, –.208 for younger adults, and.260 for older adults, respectively. Correlations in the novice sample were –.255 for children,.208 for adolescents,.133 for younger adults, and.136 for older adults. No single correlation reached the significance level.

Again, JOL and CJ differentiation was assessed by calculating mean judgments before correct versus incorrect responses (cf. Table 4), which were then compared in an ANOVA with correctness of response and monitoring measure as within-subjects factors and age group and expertise as between-subjects factors. A significant main effect of expertise, F(1, 152) = 40.09, p < .001, ηp2 = .21, indicates that experts (5.32) gave higher mean ratings than novices (4.53). Furthermore, the ANOVA revealed a significant main effect of monitoring measure, F(1, 152) = 18.65, p < .001, ηp2 = .11: In general, CJs (5.06) received higher ratings than JOLs (4.79). In addition, the main effect of correctness of response, F(1, 152) = 510.96, p < .001, ηp2 = .77, the interaction between correctness of response and age group, F(3, 152) = 3.83, p < .05, ηp2 = .07, and the triple interaction between correctness of response, age group, and expertise, F(3, 152) = 4.01, p < .01, ηp2 = .07, reached the significance level. To interpret these effects, separate ANOVAs with expertise as a between-subjects factor and correctness of response as a within-subjects factor were conducted for each age group. We found that participants in all age groups were able to differentiate between correct and incorrect responses and gave higher monitoring judgments to correct responses than to incorrect responses (all ps <.001; all ηp2 > .65). In children, the ANOVA additionally revealed a significant main effect of expertise, F(1, 38) = 25.85, p < .001, ηp2 = .41, as well as an interaction between expertise and correctness of response, F(1, 38) = 12.36, p < .01, ηp2 = .25. The latter indicates that experts (5.44) among children gave higher mean judgments than novices (4.23) and that novices in this age group (5.37 vs. 3.04) differentiated better between correct and incorrect responses than did experts (6.12 vs. 4.77). In adolescents and younger adults, no further main effect or interactions were found. In older adults, the ANOVA also revealed a significant main effect of expertise, F(1, 38) = 20.98, p < .001, ηp2 = .36, with experts (5.69) giving higher mean ratings than novices (4.64). There were no significant main effects or significant interactions.

Table 4 Mean JOLs and CJs in correct versus incorrect responses as a function of age group and expertise (Study 2)

8 Note. JOLs = judgments of learning; CJs = confidence judgments.

As in Study 1, monitoring accuracy was assessed by calculating gamma correlations between monitoring judgment and recall performance for each participant. T tests showed that all correlations displayed in Table 3 are different from 0 (JOLs, p < .01 in experts among older adults; all other ps <.001; CJs, all ps <.001). An ANOVA with age group and expertise as between-subjects factors and monitoring measure as a within-subjects factor revealed a significant main effect of monitoring measure, F(1, 151) = 8.38, p < .01, ηp2 = .05: Gamma correlations in CJs (.64) were higher than those in JOLs (.55). Furthermore, a significant interaction between age group and expertise was found, F(3, 151) = 3.71, p < .05, ηp2 = .07. T tests revealed that in children, novices (.74) had higher mean correlations than experts (.58). In the other three age groups, no differences between the two experimental groups were found. No further significant main effects or interactions were found.

The second experiment was designed to extend the findings of Study 1 to a metacomprehension approach. That is, on the one hand, Study 2 was conducted to strengthen the still poor data base related to developmental studies on the relation between metacognition and expertise in different age groups. On the other hand, it seemed important to include another experimental paradigm, because metacomprehension processes are known to be closer to more complex everyday life and therefore are not directly comparable to monitoring of item pairs (Thiede et al., [51]).

The results show that we succeeded in adjusting item difficulties across the age groups, as mean recall ranged from 0.4 to 0.7 for questions on the text. Again, there was no evidence for floor or ceiling effects, and only small age effects were found: Young adults performed slightly better than adolescents and older adults, and experts outperformed novices.

The global judgment turned out to be a rather imprecise rating: All correlations with later performance were close to 0. This finding is in line with several other outcomes stating that prediction and recall should be collected in a very similar way to ensure the diagnosticity of potential self-testing (e.g., Begg, Duft, Lalonde, Melnick, & Sanvito, [2]). So even if a global judgment measure represents a "standard" (Dunlosky et al., [13], p. 79) in metacomprehension literature, our results provide further evidence that monitoring of texts should be measured on a more detailed level.

This requirement is fulfilled in JOLs and CJs. Differentiation and accuracy turned out to be satisfactory. Participants from all age groups and levels of expertise gave higher judgments for correct than for incorrect responses. In total, ratings for CJs were somewhat higher than for JOLs regardless of the correctness of the answer. The quality of differentiation was comparable between the age groups and thus in line with the findings from Study 1 and other previous study findings (e.g., Hertzog & Dunlosky, [20]; Von Der Linden et al., [56]). The accuracy level also corresponds to findings of other metacomprehension studies, with a mean gamma correlation between JOLs and later recall of.55 and between CJs and recall of.64 (e.g., Dunlosky et al., [13]; Von Der Linden & Roebers, [55]). Again, the accuracy was higher in CJs than in JOLs reflecting the availability of even more diagnostic cues in CJs (Koriat & Goldsmith, [29]; Rhodes & Tauber, [38]).

As in the first study, the main focus was on the impact of domain knowledge on the different monitoring measures in the respective age groups. Concerning the global judgment, we could not identify consistent influences of expertise. Regarding the more specific monitoring measures (JOLs and CJs), we found a similar trend as in Study 1 with experts giving higher mean judgments regardless of the correctness of the response. However, this finding was only true for children and older adults. In children, it was additionally found that novices surpassed experts both in differentiation and in accuracy. Thus, the effect of expertise was especially pronounced for incorrect responses in third graders.

The results of the two studies presented here seem to indicate that the cognitive and metacognitive effects of domain knowledge have to be considered separately. The positive effect of expertise on memory performance, which has already been shown in several experiments (Feltovich et al., [15]; Simon & Chase, [48]), was confirmed in both studies. However, this benefit does not fully apply to metacognitive performance: Better memory did not improve monitoring per se, as other studies suggest (e.g., Griffin et al., [19]; Nietfeld & Schraw, [37]).

The inclusion of various monitoring measures within one single analysis allowed us to compare differential effects of expertise. Contrary to our expectations, we found that prior knowledge had a similar impact on monitoring at prospection and retrospection. Monitoring performance for both experimental groups was best for CJs, followed by JOLs and EOLs as well as the global judgment, which is in accord with previous findings (Dunlosky et al., [13]; Leonesio & Nelson, [31]; Von Der Linden & Roebers, [55]). While no influence of expertise was found regarding the global judgment, the expected enhanced influence of expertise on JOLs and CJs could not be confirmed. It thus seems that other cues, which might be item characteristics such as complexity and concreteness, are more influential for monitoring quality than previous knowledge (Jönsson & Lindström, [23]). As the precise interaction of different cues still remains unclear (cf. Koriat, [25], [26]), this finding could explain the diverging results concerning the relation of monitoring and expertise and seems to be a promising starting point for future research.

Furthermore, our studies demonstrated that findings concerning stimuli of different complexity (i.e., paired-associates learning vs. metacomprehension) should not be lumped together. The results do not confirm the aforementioned assumption that a metacomprehension design provides greater advantages for experts compared with novices: Whereas consistent benefits of previous knowledge were found in Study 1's gamma correlations, experts in Study 2 did not surpass novices either in differentiation performance or in accuracy. When answering questions to a soccer-related story, third-grade novices exhibited even better monitoring outcomes for both analyses; so children's metacomprehension performance seems to be rather attenuated by previous knowledge. This finding contrasts Schneider et al.'s (1989) results: When the authors compared the performance of third-, fifth-, and seventh-grade soccer experts, they found enhanced monitoring even in the youngest group. It is possible that experts in our sample did not similarly benefit from the content of the soccer text because it did not match fully with their prior knowledge; in fact, a lot of new information had to be encoded (cf. van Loon, de Bruin, van Gog, & van Merrienboer, [54]). Moreover, the fact that only in Study 1 experts of all age groups outperformed novices in monitoring may be due to more reliable self-testing in this setting than in Study 2. In the metacomprehension setting, self-tests remained presumably more ambiguous because the correct answer had to be inferred from the text. In particular, the experts among the third graders—but to a smaller extent, also experts in the other age groups—might have based their judgments mainly on their knowledge and not so much on the information provided by the text. This may have contributed to better differentiation and accuracy in novices and a general increased optimism in the expert groups, mostly in older adults (cf. Koriat & Bjork, [28]; Toth et al., [52]). Furthermore, the metacomprehension approach presumably led the experts among older adults to rely more on familiarity with the plot compared with adolescents and younger adults (cf. Shing et al., [47]). It seems that especially experts in the two middle age groups had less difficulty in distinguishing between newly acquired information and previous knowledge when memorizing information from the text. The item pairs, however, seemed to activate feelings of familiarity in experts of every age group. The close links between different domain-related concepts in experts' semantical networks might have been activated to a greater extent when learning paired associates than when reading the text (cf. Anderson, [1]).

In sum, the impact of expertise on different indicators of monitoring and in different age groups seems rather small. It is important to validate Toth et al.'s ([52]) and Schneider et al.'s ([41]) findings with stimuli with adjusted difficulty for the different age groups, as we did in our study. In older adults, Connor, Dunlosky, and Hertzog ([7]) found that age-related differences in calibration disappear when mean performance lies around 50%. This finding may be an additional reason for why we did not find a further age effect on monitoring, especially in senior experts: Baseline performance was assimilated among the age groups by means of different presentation times and texts of adjusted difficulty.

However, for all age groups, the most notable and consistent influence of expertise on monitoring performance consists in the greater optimism of experts concerning their own accomplishment—a result that applies as well for all types of stimuli and metacognitive measures. Although not explicitly mentioned in the instruction, the importance of expertise was salient in the two experiments and presumably has triggered confidence in experts (cf. Son & Kornell, [49]). Thus, expertise can be seen as an anchor that—in the course of learning and retrieval—is adjusted more or less successfully on the basis of actual experiences with the current task (Zhao & Linderholm, [58]). In addition, it is possible that experts misinterpret their more automatized processing of information (de Bruin et al., [8]; Nietfeld & Schraw, [37]) as a cue for EOLs, JOLs, and CJs, which may cause an illusion of knowing. A closer inspection of other studies indicating experts' superiority in metacognitive monitoring compared to novices reveals that it is difficult to infer to what extent participants with more domain knowledge also exhibited more optimism. For example, in the study by de Bruin et al. ([8]) experts came up with more positive predictions than novices; at the same time, the percentage of correctly predicted chess moves also exceeded that of novices. Griffin et al.'s ([19]) study on baseball-related knowledge showed less underconfidence in experts than in novices, a finding that may be equivalent to the increased optimism found in our studies.

One inevitable limitation of our approach concerns the selected domain. By choosing soccer, we had to take into account that—in opting for a naturalistic field—it is hard to identify individuals who have no related knowledge at all, given that soccer is a rather popular sport in Germany. On the other hand, we did not explicitly search for soccer experts but divided a random sample into persons with relatively more and less knowledge, using a median split procedure. Still, the comparison of mean performance in the soccer test showed substantial differences between experts and novices in each age group. So even if in the two older age groups, novices had some basic soccer knowledge, experts surpassed them clearly. Furthermore, our findings showed substantial and consistent differences in memory performance between our expert and novice groups. A second limitation concerns the degree of expertise in the different age groups. The fact that the amount of acquired expert knowledge changes throughout the course of life made it necessary to use different test versions for children on the one hand and the remaining three age groups on the other hand. A relatively easy soccer test was used for the youngest age group to avoid floor effects, and a more difficult version was given to the older age groups. Of course, this has an impact on the measure to use a median-split procedure in each age group to define experts and novices. Thus, we acknowledge the fact that the difference in knowledge between soccer experts and novices may be more extreme in one age group, as compared with another. However, preliminary analyses for the three oldest age groups who had received the same soccer test indicated that knowledge differences between experts and novices were roughly comparable at least in these age groups. Furthermore, the results of both studies were validated by analyses with a narrower definition of expertise that revealed the same patterns and thus emphasized the reliability of our findings.

Taken together, the results of the two studies indicate that experts judge their own performance more optimistically than novices and that more accurate or differentiated monitoring is less pronounced. Increased monitoring judgments in participants with greater soccer knowledge were consistently found for the memorizing of word pairs and for the answers to questions on a narrative. Further research should aim to replicate this finding in a larger sample to analyze the degree of overestimation by an analysis of calibration. Interestingly, the experts' generally more positive judgments were more accurate in the paired-associates design. As our findings represent the first approach using a life-span sample, our conclusions remain preliminary and require replications in other domains of expertise. Studies relating metacognitive knowledge and domain knowledge are still scarce, particularly for elementary school children and older adults. This lack of studies is surprising given that the expert–novice paradigm has a relatively long tradition. Hence, expanding our findings to more age groups would account for more fine-grained developmental changes, especially from childhood to adolescence and from younger to older adulthood. Overall, the results suggest that, independent of age, rich domain knowledge can lead to more accurate metacognitive monitoring. Additionally, our research findings indicate that the experts' more positive self-assessments reflect more optimistic judgments in many cases. However, such an optimism regarding their own achievement potential does not necessarily have negative consequences on memory performance given that experts outperformed novices on all occasions.

We wish to thank all participating children, adolescents, and adults as well as teachers, principals, and parents for their cooperation.

This research was conducted as part of a research project on the development of procedural metacognitive knowledge across the life span and was financed by the German Research Foundation (DFG-Gz. SCHN 315/45–1).