Knowledge exploration encompasses the active pursuit of information, ideas, and experiences, extending beyond the boundaries of immediate task demands. This proactive behaviour arises from various motivational states, such as curiosity, interest, and engagement. Exploration plays a vital role in driving advancements in our society, serving as a fundamental catalyst for learning as well as new discoveries. Consider the situation of a student who cannot provide an answer to a question during an exam. The student’s recognition of the lack of knowledge will only become valuable when they actively delve into exploring and acquiring knowledge about the specific topic.

Scholars have operationalised knowledge exploration as the dynamic process of seeking out and uncovering new information (Liu et al., 2010). This ongoing and dynamic process governs how individuals interact with and learn from their environments (Hardy et al., 2014; Hacques et al., 2021). A broad spectrum of activities, including searching for, generating, and responding to novel opportunities, constitutes “exploratory activities” (Kraiprasit & Bhanthumnavin, 2022). Other forms of exploratory behaviour, as identified by Berlyne and Lewis (1963) include fact-finding responses, posing queries, and consulting written sources. The impetus for exploration is often triggered by recognizing gaps in understanding, representing epistemic curiosity, which is an internal emotional drive to acquire knowledge (Litman & Spielberger, 2003). This drive prompts individuals to actively seek information until a satisfactory level of understanding is attained (Sawant, 2015). However, not all curiosity translates into exploratory behaviour. Specific factors trigger exploratory behaviour, such as when epistemic curiosity surpasses a certain threshold, motivating individuals to engage in active exploration. Therefore, while epistemic curiosity triggers the desire to learn, knowledge exploration refers to goal-directed actions taken to fulfil that desire.

Exploratory behaviour is highly relevant across various disciplines; however, there is a notable scarcity of behavioural measures for exploration (Gross et al., 2020). Most research has concentrated on visual and perceptual domains (Wittmann et al., 2008), which restricts our understanding of exploration across different knowledge domains (Trevors et al., 2017; Yan et al., 2013). To address this limitation, our study uses a knowledge gap-induced task to investigate real-time exploratory behaviour among student population. It contributes to the existing literature by identifying specific factors that trigger or hinder information-seeking behaviour. In this study, knowledge exploration is operationalised as information-seeking behaviour by learners aimed at enhancing current knowledge.

The majority of research on exploratory behaviour considers stimulus parameters such as complexity, novelty, and ambiguity (Schomaker & Meeter, 2015; Vogl et al., 2019), and focuses on visual or perceptual forms of exploration (Wittmann et al., 2008; Duzel et al., 2010). For instance, Berlyne and Lewis (1963) found that stimulus properties such as novelty, surprisingness, complexity, and incongruity improve our perception, enhance motivation, trigger exploratory behaviour, and foster learning (Schomaker & Meeter, 2015). In addition to stimulus-related aspects, research on dispositional factors associated with exploration has only recently begun. For instance, Kraiprasit and Bhanthumnavin (2022) found that psychological states predict knowledge exploration, while psychological traits indirectly influence exploration. To further our understanding of how dispositional factors influence knowledge exploration, this study examines the role of intrinsic and extrinsic motivation.

Beyond dispositional variables, contextual factors also significantly impact learning (Wang et al., 2019). For instance, feedback aids students in adjusting their strategies for task completion, thereby enhancing learning outcomes (Haddara & Rahnev, 2022). In this study, we aim to delineate the role of contextual factors such as Feeling of Confidence (FOC), confidence error, and feedback in determining knowledge exploration behaviour. By integrating both dispositional and contextual variables, our study seeks to provide a comprehensive understanding of the mechanisms underlying knowledge exploration.

This paper presents the findings of a study that attempted to develop a comprehensive understanding of students’ knowledge exploratory behaviour and decipher the role of feedback, metacognitive Feeling Of Confidence (FOC), and academic motivation in explaining knowledge exploration. We used following research questions to direct our investigation:

  1. 1.

    How does the presence of feedback in a low-stake learning environment affect students’ real-time knowledge exploration?

  2. 2.

    How do the feeling of confidence and confidence error influence real-time knowledge exploration?

  3. 3.

    What is the role of intrinsic and extrinsic academic motivation on students’ real-time knowledge exploration?

Factors affecting knowledge exploration

Feedback

It is believed that the presence or identification of collative variables serves as the catalyst for the increased awareness and attention associated with knowledge seeking. In any academic setting, students often get one or more of three sorts of response on their work: a grade, a statement of adulation or criticism, and some details of their performance. Students frequently use the response they receive to get an overview of how they did and advice on how to do better (Lipnevich & Smith, 2009). The information received about the discrepancy between actual and desired level of performance is what we call feedback, which is an important precursor to learning and achievement (Hattie & Timperley, 2007). In simpler words, feedback is the information provided by an agent regarding one’s quality of performance. The agent could be a teacher, family, peers, self, or computer-generated information. Getting feedback helps a learner confirm, expand, improve, or restructure information. Providing feedback to students plays a crucial role in motivating them to continue positively with their learning journey. Although feedback has been considered important in learning, a limited number of studies have investigated the meaning of feedback in teaching space (Hattie & Timperley, 2007). Goodman et al. (2004) suggested that when feedback is given immediately after incorrect behaviour, the learning is highest. Murayama (2022) suggested that when students are given negative feedback about their performance, it decreases their propensity to seek further information. Feedback in the form of verbal compliments has been found to boost interest by elevating one’s sense of competence, but feedback in the form of financial rewards for task completion has been proven to lower intrinsic motivation for a variety of activities (Singh & Manjaly, 2021).

Few theorists argue that feedback acts as a reinforcement to enhance behaviour and learning, making it a direct precursor to learning (Wang et al., 2019), while some believe that feedback is a cue to make people modify their strategy and thus have an indirect effect on learning (Vollmeyer & Rheinberg, 2006). Haddara and Rahnev (2022) conducted a study to understand the mechanism of feedback in a perceptual decision-making task. They provided feedback to one group of participants after every trial and withheld feedback from the other group. It was observed that the trial-by-trial feedback helps students in modifying the strategy for completing the task and in improving confidence calibration.

Therefore, feedback acts as one of the most pervasive and effective strategies to change behaviour across a wide range of contexts, including academic, professional, and recreational settings as well as a variety of cognitive activities (Hattie & Timperley, 2007). Its capacity to enhance performance is well-documented, yet the extent and consistency of its impact across different domains remain subjects of considerable debate (Williams & Williams, 2022). While certain studies within the realm of perceptual tasks have demonstrated that feedback can significantly improve performance (Cavalcanti et al., 2021), others have reported no significant difference in performance outcomes between groups receiving feedback and those that do not (Khan & Pardo, 2016; Jin, 2017). This discrepancy highlights the need for more comprehensive research that spans a broader spectrum of fields, accounts for variations in institutional settings (Dawson et al., 2019). The necessity for further investigation into how feedback specifically impacts student learning is thus evident (Leong et al., 2017).

Given the complexities surrounding the efficacy of feedback in educational settings, particularly when considering the predominance of research centered around reward-based learning mechanisms, we aim to delve into how students navigate feedback within low-stakes learning environments. This interest stems from an acknowledgment of the subtle role feedback may play in these settings, potentially differing significantly from its impact in high-stakes or reward-focused contexts. This inquiry is grounded in the premise that understanding the role of feedback in such environments could offer valuable insights into optimizing learning experiences. To systematically explore this area, the following hypothesis was proposed in line with the existing literature on the role of feedback in enhancing performance:

H1

The presence of accuracy feedback in the task positively influences knowledge exploration.

Feeling of confidence

Feeling of Confidence (FOC) represents an individual’s belief in the correctness of their answers (Efklides, 2009). It serves as a pivotal measure of metacognitive feelings within various cognitive tasks, including eyewitness testimony and general knowledge evaluations (Efklides, 2002). This subjective certainty is not merely an assessment of probability but is deeply intertwined with the outcomes of cognitive processing, reflecting the degree to which solutions or responses are perceived as accurate (Efklides, 2009).

In cognitive psychology, FOC is primarily considered retrospective, emerging after the completion of a task (Efklides, 2002, 2006). These feelings lack a deliberate analytical foundation but are crucial in providing intuitive feedback on performance and play a significant role in signalling personal engagement with the task (Efklides, 2009). This intuitive nature distinguishes FOC from other metacognitive judgments like judgment of solution correctness, which is driven by explicit, task-specific information and involve more reflective and analytical cognitive strategies. Metacognitive judgments allow individuals to plan, monitor, and adjust their strategies in alignment with the precise demands of the task (Efklides & Petkaki, 2005). Despite the established significance of metacognition in enhancing learning outcomes, research into this aspect of metacognitive experiences has been relatively limited in scope (Tay et al., 2024).

Understanding the levels of confidence and their effects remains a critical concern for educators and researchers alike, as it significantly influences students’ affective and cognitive experiences. Asik and Erktin (2019) explored this by assessing metacognitive experiences, such as the feeling of familiarity, and confidence and their relationship to performance in math problem-solving tasks. They found a significant correlation, with successful students exhibiting higher feeling of confidence compared to their less successful counterparts. Dindar et al. (2020) investigated the intricate relationship between learning results and metacognitive experiences in collaborative problem-solving environments. It was found that the group’s perceived performance was significantly associated with their feeling of confidence and their objective performance was significantly predicted by the high feeling of confidence.

The metacognitive FOC significantly influences the way students navigate their learning journeys, often serving as an internal compass that directs their engagement with tasks and assimilation of new information. Generally, this sense of confidence is elevated when students are dealing with familiar tasks or content (Efklides, 2011), reinforcing their commitment to the learning process. However, a critical turning point emerges when students confront situations where their pre-existing knowledge is flawed, or when they encounter new information that starkly contrasts with their prior beliefs or understandings. Such instances precipitate a state of cognitive dissonance, as delineated by Vogl et al. (2019). This cognitive dissonance is not merely a roadblock but rather a pivotal moment in the learning process, setting the stage for what is known as the hypercorrection effect. The hypercorrection effect posits that individuals are more likely to correct their misconceptions with a greater degree of accuracy and retention when they initially hold those misconceptions with high confidence (Butler et al., 2011). When faced with the feedback that their strongly held beliefs are incorrect, the resultant surprise or shock is believed to enhance the encoding of correct information, making the learning episode particularly memorable (Metcalfe & Finn, 2012). Two key drivers have been identified for the pronounced hypercorrection effect. Firstly, the experience of surprise and embarrassment upon discovering their errors motivates people to concentrate their attention on remembering the correct information more accurately (Metcalfe & Finn, 2012). Secondly, their semantic understanding related to the task at hand tends to be stronger in instances where they make errors with high confidence as opposed to when their errors are accompanied by low confidence (Butterfield & Mangels, 2003).

While the hypercorrection effect has been extensively studied in laboratory contexts, much of this research has focused on non-academic tasks (Metcalfe & Miele, 2014; Carpenter et al., 2018; Vogl et al., 2019). This emphasis on non-academic settings has left a noticeable gap in understanding the phenomenon within educational environments. Researchers have highlighted the need to explore the hypercorrection effect using academic or educational content to bridge this divide (Vogl et al., 2019). Van Loon et al. (2015) undertook a study utilizing content from the science curriculum to investigate the hypercorrection effect among college students. Contrary to expectations and existing literature on the hypercorrection effect, their findings did not reveal any significant presence of the effect within this context. A subsequent study conducted by Carpenter et al. (2018) took a closer look at the hypercorrection effect within an academic setting. The results demonstrated that students with a higher prior knowledge of the material tended to exhibit greater confidence in their answers. This increased confidence, in turn, led to a more effective correction of errors. These findings contribute to our understanding of the hypercorrection effect in academic learning environments, suggesting that the interplay between feeling of confidence and feedback about accuracy is more subtle than previously thought.

Building upon the existing literature, which predominantly highlight the positive correlation between confidence and task success, this research takes a unique approach by examining the impact of FOC and confidence error on real-time knowledge exploration behaviour of students. We anticipate that students experiencing heightened feelings of confidence demonstrate an increased propensity for engaging in knowledge exploration during tasks. Based on hypercorrection effect, we also anticipate that instances of confidence error, wherein individuals misjudge their own confidence, may also influence students to engage more extensively in knowledge exploration behaviour. Through an empirical investigation using tasks derived from standard academic content, our study aims to elucidate the role of FOC and confidence error in predicting students’ real-time knowledge exploration behaviours. We propose the following hypotheses to systematically understand the intricate interplay between FOC and the knowledge exploration behaviour.

H2

High FOC positively predicts knowledge exploration behaviour among students.

H3

Confidence error positively predicts knowledge exploration behaviour among students.

Academic motivation

Academic motivation refers to the internal and external factors that drive individuals to engage in learning and academics-related activities. Academic motivation has been extensively studied in educational contexts (Clark et al., 2014; Hidajat et al., 2020; Rienties et al., 2009). It serves as a driving force influencing learners’ attitudes and learning behaviour (Rienties et al., 2009; Ryan & Deci, 2000) and is considered a key determinant of academic achievement (Vecchione et al., 2014). Previous research has shown that academic motivation plays a significant role in various educational outcomes, such as self-efficacy, success rates, persistence, and self-regulated learning strategies (Koyuncuoglu, 2021) as well as learning and performance in the classroom (Hidajat et al., 2020; Ozer & Schwartz 2019; Ross et al., 2016; Sides & Cuevas, 2020).

Academic motivation is commonly categorised into intrinsic and extrinsic motivation (Ryan & Deci, 2000). Intrinsic motivation drives students to study for the sake of pleasure and fulfilment. Studies have argued that our actual competencies in the academic area are developed when the purpose behind learning is internal choice and pleasure, i.e., when the source of motivation is intrinsic (Ross et al., 2016). Various mechanisms have been proposed to suggest why intrinsic motivation leads to higher academic outcomes. For instance, a study on first-year undergraduate students (Clark et al., 2014) discovered that intrinsic motivation enhances academic integration, resulting in improved academic performance. When students are intrinsically motivated to learn and perform, their academic performance improves, leading to increased persistence and goal achievement.

When individuals engage in an activity to receive external rewards or avoid punishment, they are considered to be externally regulated, i.e. their motivation is extrinsic (Clark et al., 2014; Erten, 2014; Yan et al., 2013). Extrinsically motivated students rely on external rewards to initiate their efforts, and the extent of their effort is influenced by the anticipated value of the reward. Research has shown that these students often lack persistence and are frequently unable to achieve their goals (Amrai et al., 2011; Ross et al., 2016; Vecchione et al., 2014; Yang et al., 2006).

Extensive research has explored academic motivation and educational outcomes (Tohidi & Jabbari, 2012). However, there is limited research examining the factors that influence real-time knowledge exploration (Hardy et al., 2014). This gap was also highlighted by (Dubnjakovic, 2017), who stated, “Information-seeking motivation has garnered little attention from researchers” (p. 1034). Given that motivation is linked to various academic outcomes, such as achievement and enhancement in learning (Gupta & Mili, 2017; Amrai et al., 2011; Widodo et al., 2018), this study investigates the role of academic motivation (both intrinsic and extrinsic) as a predictor of knowledge exploration behaviour. Therefore, we formulated the following hypotheses:

H4

Intrinsic motivation positively predicts knowledge exploration behaviour.

H5

Extrinsic motivation negatively predicts knowledge exploration behaviour.

Data and methods

The purpose of the study was to understand real-time knowledge exploration among high school students. We used a low-stake experimental task-based paradigm where participants were exposed to a variety of questions in two sessions of 15 trials each. We explicitly remarked that there was no performance judgment associated with the task. Figures 1 and 2 present the experimental framework for sessions 1 and 2, respectively. The framework was implemented in E-Prime software. In session 1, multiple-choice questions were presented in random order, and respondents were asked to choose the correct answer. The participants had the option to choose between three options: (a) move to the next question indicating no motivation to explore; (b) see the correct answer; (c) see the detailed answer indicating real-time knowledge exploration. When the participants chose the third option, they were presented with a detailed description of the topic on a single screen where they could learn about the topic and the rationale for the correct answer. Session 2 was similar to Session 1 with two additions; immediately after the response, participants were asked to indicate their confidence in the answer, followed by performance feedback.

Fig. 1
figure 1

Experimental framework for session 1

Full size image
Fig. 2
figure 2

Experimental framework for session 2

Full size image

Participants

Using snowball sampling, school-going students of classes 9 and 10 (commonly known as high school) were requested to participate in the study voluntarily. 100 high school students (48 female, 52 males) participated in the study (age range = 13–17 years, Mean = 14.7 years, SD = 0.78). The informed consent of the volunteering students and their parents was obtained.

Procedure

The experiments were conducted in a quiet room resembling a laboratory environment within the school/ study centre premises. Students were first briefed about the study and were assured of anonymity and confidentiality. They were informed that the study would take about one hour and be conducted in two sessions. Before participating in the experimental study, they filled out an online questionnaire on academic motivation (Vallerand et al. 1993). The first session followed the framework presented in Fig. 1. Multiple choice questions (MCQs) with four options were presented in fifteen trials. Students were instructed to choose an option that seemed correct. After making their decision, they were not provided feedback about the accuracy. After this, they had the option to see the correct answer, skip seeing the answer, or choose to see the detailed answer.

On the same day, students were again called for the second session. The second session followed the framework presented in Fig. 2. The second set of MCQs was presented in a randomized order on the computer screen. As shown in Fig. 2, two amendments were made in the procedure of session 1. First, the participants were asked to rate their feeling of confidence on a 5-point Likert scale (1 = not at all confident to 5 = very confident) in each question and thereafter, immediate performance feedback about the accuracy of their response flashed on the computer screen (Your answer is incorrect; Your answer is correct). After this, they had the option to see the correct answer, skip seeing the answer, or choose to see the detailed answer. The study was approved by the Human Research Ethics Committee of the institute.

Materials

Experimental Task

The task used in the sessions 1 and 2 consisted of two sets of 15 multiple-choice questions compiled from standard high school textbooks. All the questions were chosen with the help of three high school teachers, and items were specific to the educational level of class 9 and 10 students. The questions covered various topics such as mathematics, biology, science, and general knowledge. We included questions from different domains to understand knowledge exploration from an unbiased viewpoint. Notably, the range of item difficulty was similar across both sessions.

Knowledge Exploration

At the end of each trial, after answering the question, participants were given three choices: skip directly to the next question (indicating no exploration), view only the correct answers (indicating minimal exploration), or select to see the detailed answer (indicating detailed exploration).

FOC

Following each trial in session 2, after participants answered the question, they were asked to rate their confidence in the correctness of their responses before receiving feedback. This FOC was measured using a 5-point Likert scale. Participants responded to the question, “How confident are you that your answer is correct?” with options ranging from 1 (“not at all confident”) to 5 (“very confident”).

Academic-motivation-scale (AMS)

Academic-Motivation-Scale (Vallerand et al. 1993), consisting of 28 items, was used to measure students’ extrinsic and intrinsic motivation for learning on a seven-point Likert scale (1 = very untrue of me to 7 = Very true of me).

Analytical Plan

The data are organised hierarchically into two levels, with questions at Level 1 [L1] nested within Individuals (Level 2 [L2]). Using Mplus 8 (Muthén & Muthén, 2017), we modelled within- and between-person relations in these nested data using multilevel modelling. Our sample included 100 participants on level 2, with 15 trials on level 1. This sample size fits with the recommendations by Arend & Schäfer (2019) for two-level models to detect small, medium, and large L1 effects. The outcome variable knowledge exploration has three ordered response categories, thus representing an ordered categorical response variable (Heck & Thomas, 2020). Therefore, we conducted multilevel regression analysis with an ordinal logit model to estimate the relationship between knowledge exploration (dependent variable) and three within-level variables – accuracy, FOC, and accuracy*FOC (L1 predictor variables) and two between-level variables – intrinsic motivation and extrinsic motivation (L2 predictor variables). Before conducting the analyses, we examined the data for the suitability of multilevel analysis by calculating the Intra-Class Correlation (ICC) for our outcome variable (knowledge exploration). This is a crucial step in determining whether to proceed with multilevel analysis (Hox et al. 2017). ICC indicates the degree to which the multilevel data structure might impact the outcome variable of interest. It describes the relative strength of the clustering present among the groups, and it is expected to be close to zero (McNeish & Stapleton, 2016). ICC value of our data supported the usage of multilevel modelling (Heck & Thomas, 2020).

To analyse Hypothesis 1, we integrated the data from sessions 1 (Absence of feedback) and 2 (Presence of Feedback) to examine the impact of feedback on knowledge exploration behaviours among students in a low stake learning environment. We transformed our ordinal outcome variable, knowledge exploration, into a binary format (0 representing ‘No exploration’ and ‘Minimal exploration’; 1 representing ‘Detailed exploration’) to facilitate a clearer interpretation and simplify the modelling of exploratory behaviour. We employed a within-subjects logistic regression to assess the influence of feedback (0 = no feedback, 1 = feedback) on knowledge exploration behaviours (Table 1).

Table 1 Mean, SD, Intra-class correlation and Pearson correlation between variables for both the sessions
Full size table

To investigate the role of FOC and confidence error in knowledge exploration behaviour (Hypotheses 2 & 3 respectively), we modelled accuracy, FOC, and the interaction term accuracy*FOC as predictors of knowledge exploration at L1 using data from session 2. Accuracy and FOC were standardized before calculating the interaction term. Group mean centering of predictors was done to disentangle within and between-person effects. The inclusion of the interaction term accuracy*FOC aimed to elucidate the impact of confidence judgments on exploration behaviours in the context of both correct and incorrect responses.

For hypothesis 4, level 2 variables (intrinsic motivation and extrinsic motivation) were added in the model. Intrinsic and extrinsic motivation were first grand mean centered and then modelled as predictors at L2 (Model 2). We assessed the goodness of the model fit by comparing the Akaike Information Criterion (AIC) values derived from the models, where a lower AIC value denotes a more appropriate fit (Finch & Bolin, 2017; Heck & Thomas, 2020). The ‘marginsplot’ command in Stata 15 was utilized to graphically represent the results, for facilitating a deeper understanding through the visual illustration of how the predictors influence the exploratory behaviours of students.

Result

We first present the descriptive statistics for the variables. The results of multilevel regression analysis are then presented to determine the factors that predict knowledge exploration among students.

Table 2 shows descriptive statistics (Mean, SD, ICC, Pearson correlation) for variables studied in session 1 and session 2. ICC (Intra Class Correlation) for knowledge exploration in the two sessions were 0.544 and 0.271, respectively, which suggested significant variation at within person level, thus supporting the usage of a multilevel regression model.

Table 2 Standardized coefficients for feedback as a predictor of knowledge exploration
Full size table

Feedback and knowledge exploration (hypothesis 1)

Our first hypothesis (H1) posited that the presence of accuracy feedback in the task would positively influence knowledge exploration tendencies. To assess the significance of feedback, we applied logistic regression model using feedback as a predictor of detailed knowledge exploration (Table 1). The results demonstrate a significant effect of feedback on detailed knowledge exploration behaviour. Specifically, when comparing session 1, where feedback was not given to students after each trial, to session 2, where feedback was presented to students after each trial, a positive effect was observed (β = 0.270, SE = 0.07,p < .001). Furthermore, the odds ratio associated with this effect was calculated to be 1.31, suggesting that the presence of feedback increases the likelihood of detailed knowledge exploration behaviour among students by approximately 31%. Thus, our findings highlight the significant positive influence of feedback (knowledge of accuracy) on detailed knowledge exploration behaviour among students, supporting our hypothesis (H1).

Figure 3 shows the graphical representation of the relationship between feedback and the probability of detailed knowledge exploration tendencies. The x-axis represents feedback condition (0 = No feedback; 1 = feedback present). The y- axis denotes the probability of detailed knowledge exploration. From the graph, it is evident that presence of feedback in trials enhanced the probability to engage in knowledge exploration behaviour among students. Whereas absence of feedback reduced the probability to engage in knowledge exploration behaviour among them.

Fig. 3
figure 3

Graph showing probability of detailed knowledge exploration with respect to feedback

Full size image

Feeling of confidence (FOC) and Knowledge Exploration (hypothesis 2 & 3)

Hypothesis 2 and 3 posited that students experiencing heightened feelings of confidence would demonstrate an increased propensity for engaging in knowledge exploration, and the instances of confidence error would also influence students to engage more extensively in knowledge exploration behaviour respectively. Model 1 (Table 3) demonstrates a reasonably good fit, evidenced by improvement in the AIC value from the unconditional model (AIC = 3012.35) to model 1 (AIC = 2311.47). The findings presented in Model 1 (Table 3) indicate that accuracy negatively predicts knowledge exploration (β= -0.682, SE = 0.020,p < .001), suggesting that the knowledge exploration behaviour increased significantly for incorrect answers. FOC positively predicts knowledge exploration behaviour (β = 0.068, SE = 0.025,p < .001), in line with our hypothesized relationship, suggesting that knowledge exploration increases with increase in FOC (H2). The interaction between accuracy and FOC (accuracy*FOC) (β= -0.179, SE = 0.028,p < .001) emerges as a significant negative predictor of knowledge exploration, highlighting the interplay between confidence and accuracy. This interaction effect suggests that the relationship between FOC and exploration is contingent upon the correctness of the response. Specifically, confidence error, i.e., high confidence in incorrect answers, serves as a robust predictor of increased knowledge exploration, thlous supporting our hypothesis (H3) regarding the positive impact of confidence errors on exploratory behaviour.

Table 3 Multilevel ordinal logit regression models for explaining knowledge exploration
Full size table
Fig. 4
figure 4

Figure showing FOC*ACC as a predictor of knowledge exploration

Full size image

Figure 4 demonstrates the interaction between the FOC and accuracy (ACC) in influencing knowledge exploration levels. The probability of not engaging in exploration at all increased significantly for answers that were correct and for which confidence was high (ACC = 1, Outcome = 0). Also, as the feeling of confidence increased, the probability of engaging in detailed exploration decreased for correct answers (ACC = 1, Outcome = 2). Whereas for the incorrect answer given with high confidence, the probability for detailed exploration increased steadily (ACC = 0, Outcome = 2), thereby depicting the occurrence of a hypercorrection effect. Overall, this graph highlights how the interplay of accuracy and the FOC together influence the tendencies towards varying depths of knowledge exploration.

Motivation and knowledge exploration (hypothesis 4 & 5)

Our next hypotheses posited that intrinsic motivation would positively predict knowledge exploration (H4) while extrinsic motivation negatively predicts knowledge exploration (H5). Model 2 (Table 3) introduced the Level 2 variables to predict knowledge exploration. Intrinsic motivation and extrinsic motivation were first grand mean-centred and then modelled as predictors of knowledge exploration. Level 1 variables (accuracy, FOC, accuracy*FOC) were retained in building the current model. The final contextual model (Model 2) in Table 3 consisted of intrinsic and extrinsic motivation at level 2 (between level) and accuracy, FOC, accuracy*FOC at level 1 (within level). The AIC value of model 2 decreased compared to previous models, suggesting a reasonably good model fit. The results indicate that the addition of between-level variables improved the prediction of knowledge exploration above and beyond the within-level variables included in the model. The final model suggests that intrinsic motivation positively predicted knowledge exploration (β = 0.389, SE = 0.087,p = .001). The positive coefficient suggests that higher levels of intrinsic motivation are associated with increased tendencies of knowledge exploration, thus supporting our hypothesis (H4). Interestingly, extrinsic motivation negatively predicted knowledge exploration (β =-0.339, SE = 0.088,p = .001), suggesting that a higher level of extrinsic motivation is associated with decreased tendencies of knowledge exploration, thus supporting our hypothesis (H5).

Fig. 5
figure 5

Figure showing intrinsic motivation as a predictor of knowledge exploration

Full size image
Fig. 6
figure 6

Figure showing extrinsic motivation as a predictor of knowledge exploration

Full size image

Figure 5 presents the relationship between intrinsic motivation and the likelihood of knowledge exploration. The X-axis represents intrinsic motivation levels, ranging from 19 (low intrinsic motivation) to 79 (high intrinsic motivation), and the Y-axis represents the probability of knowledge exploration, ranging from 0 to 1. From the graph, it can be observed that as intrinsic motivation increases, the probability of engaging in no exploration decreases sharply, while the probability of detailed exploration (outcome = 2) increases with an increase in intrinsic motivation. Overall, the graph highlights the influence of intrinsic motivation on detailed knowledge exploration: those students having high intrinsic motivation showed an increased likelihood of detailed knowledge exploration, whereas those students with low intrinsic motivation were less likely to engage in knowledge exploration behaviour and more likely to show non-exploration.

Figure 6 presents the relationship between extrinsic motivation and the likelihood of knowledge exploration. The X-axis represents extrinsic motivation levels, ranging from 27 (low extrinsic motivation) to 81 (high extrinsic motivation) and the Y-axis represents the probability of knowledge exploration behaviour, ranging from 0 to 1. The graph indicates that as extrinsic motivation increases, the probability of engaging in no exploration (Outcome = 0) increases sharply, while the probability of detailed exploration (Outcome = 2) decreases significantly with an increase in extrinsic motivation. Overall, the graph highlights the influence of extrinsic motivation on knowledge exploration: those students having high extrinsic motivation showed a decreased likelihood of knowledge exploration and were more likely to show non-exploration tendencies, and students with low extrinsic motivation were more likely to engage in detailed knowledge exploration.

Discussion

The present study provides significant insights into our understanding of knowledge exploration. It focuses on students’ behaviour in a low-stakes learning environment and seeks to shed light on the interplay between feedback, metacognitive feeling of confidence, and academic motivation.

As expected, the presence of feedback in the task positively influenced students’ engagement in knowledge exploration within the task (H1). Our results are consistent with the prior research emphasizing the importance of feedback in enhancing learning outcomes and performance. A study by Tricomi and DePasque (2016) highlighted how feedback on their responses not only informs students but also motivates them toward further learning. Feedback is recognized for its dual function of providing information and fostering motivation in the learning process. Hattie and Timperley (2007) also reinforced the importance of precise and timely feedback in student learning and achievement. Feedback helped students develop effective self-regulated learning strategies, error correction, and promoting deeper learning (Banerjee et al., 2020; Cavalcanti et al., 2021; Thomsen et al., 2022; Wang et al., 2019). Our findings also corroborate with the previous research conducted in online learning environments, affirming that feedback improves student performance (Cavanaugh, 2013), enhances student engagement and motivation compared to those without feedback (Kwon et al., 2017), thereby promoting cognitive engagement and learning (Wu & Schunn, 2020). While existing studies have predominantly focused on evaluating student performance and achievement in relation to feedback, we have utilized feedback to understand students’ willingness to seek explanations and bridge their knowledge gaps in low-stakes learning environments (i.e. conditions devoid of monetary or academic incentives).

For hypothesis 2, the findings indicate that the high feeling of confidence positively influenced the knowledge exploration behaviour. When students reported high confidence in their responses, they were more likely to explore the knowledge and choose detailed exploration. This could be because when students are confident in their knowledge, they are more likely to recognize their own limitations and areas where their understanding is lacking, which in turn motivates them to delve deeper into the material. In line with this finding, Veenman et al. (2006) reported that high confidence is associated with better learning capabilities and higher achievement. They suggested that confidence acts as a facilitator for engaging more deeply with the learning material. Efklides (2011) observed that students tend to experience an increase in confidence when interacting with familiar tasks or content, which in turn enhances their dedication to the learning process. Finn and Tauber (2015) observed that confidence enhances engagement in learning tasks, potentially leading to more effective and thorough exploration of new material. Similarly, Wang (2015) found that confidence is positively associated with learning, suggesting that confident students are more likely to acquire new knowledge effectively. Overall, the literature consistently highlights the beneficial impact of confidence on several educational metrics, including enhanced performance and deeper engagement with the learning task.

As expected in hypothesis 3, the confidence error positively predicted knowledge exploration behaviour among students. It indicated that the high confidence in incorrect answers was associated with higher knowledge exploration behaviour. The result is consistent with previous studies suggesting that errors can promote deeper processing of information and improve long-term retention. A study conducted by Bjork et al. (2013) highlighted that encountering errors facilitates a deeper engagement with the material, thereby facilitating subsequent learning. Metcalfe et al. (2020) observed that individuals exhibit a heightened ability to learn from their mistakes when they hold confidence in their erroneous responses. Koriat et al. (2000) proposed the self-consistency model of subjective confidence, which showed that learners were more likely to correct their errors when they were confident in their initial answers. Such confidence-related errors are instrumental in enhancing metacognitive awareness and cultivating a more efficacious learning environment. These errors are reflective of cognitive dissonance as the incoming information is not consistent with the beliefs or information we hold. Thus, our finding provides empirical evidence for the hypercorrection effect. Our results confirm that confidence errors are a form of beneficial error that can enhance learning by promoting deeper processing. Our study contributes to the expanding body of research on the relationship between confidence and exploration, which has been the subject of considerable interest in recent years (Vogl et al., 2019).

For hypotheses 4 and 5, our findings were in line with the expectations, demonstrating that intrinsic motivation (H4) and extrinsic motivation (H5) were significant predictors of knowledge exploration. The findings indicated that students who exhibited higher intrinsic motivation were more likely to opt for detailed exploration in the task, while the students who exhibited higher extrinsic motivation were less likely to choose detailed exploration in the task. Intrinsic motivation positively predicted knowledge exploration, consistent with empirical findings that intrinsically motivated students exhibit a zeal for learning and performance (Clark et al., 2014; Erten, 2014; Hidajat et al., 2020; Ozer & Schwartz, 2019; Ross et al., 2016). Dubnjakovic (2017) observed that intrinsically motivated students seek out novelty through natural curiosity and exploration tendencies and they derive a sense of satisfaction and reward from the activity itself. This might explain our finding that students high on intrinsic motivation tend to delve deeper into the task at hand and explore it in detail, even in the absence of any rewards or stakes associated with it.

Extrinsic motivation, on the other hand, exhibited a negative association with knowledge exploration (H5). As our study did not involve any external incentives for exploring knowledge, students driven by external rewards may lack the motivation to engage in knowledge exploration when such rewards are absent. Previous studies have consistently demonstrated that extrinsic motivation can lead to poorer academic performance and decreased persistence (Ryan & Deci, 2000). Ross et al. (2016) also revealed that self-determination and autonomous motivation are associated with better academic achievement and persistence, while those externally regulated have less persistence and struggle to achieve their goals (Vecchione et al., 2014). Thus, extrinsically motivated students require an external reward to get started; their efforts depend on the value of the reward expected from that activity, and they lack persistence and are frequently unable to achieve goals (Ryan & Deci, 2008). This explains our finding that extrinsic motivation deters students from engaging in detailed knowledge exploration.

Conclusion, limitation and future direction

In conclusion, this study provides valuable insights into knowledge exploration among students. It significantly enriches our understanding of how feedback, FOC, and academic motivation influence students’ behaviour in low-stake learning environments. The study confirms the critical role of feedback in enhancing student engagement with the task, providing not only informative but also motivational benefits that encourage students towards exploration. Our findings also highlight the positive impact of FOC on knowledge exploration, suggesting that when students feel confident about their knowledge, they are more inclined to engage deeply with the task. Furthermore, our results provide empirical evidence to the hypercorrection effect, suggesting that confidence errors are indeed beneficial for enhancing engagement with the learning task. This study also addresses the gap on motivation and knowledge exploration, an area that has not been extensively investigated in the literature (Hardy et al., 2014). Intrinsic motivation positively predicted knowledge exploration behaviour, as intrinsically motivated students exhibit curiosity, enthusiasm, and satisfaction derived from the learning process itself. However, extrinsic motivation negatively predicted knowledge exploration, as extrinsically motivated students are driven by external rewards associated with the learning task. The findings of this study have substantial implications for educational practices and policies, highlighting the critical role of knowledge exploration in optimizing learning outcomes. This study emphasizes the importance of fostering supportive learning environments that encourage students to embrace mistakes as integral to the learning process. It is essential for educators to cultivate classroom atmospheres where errors are viewed as natural and constructive opportunities for growth. Such conducive environments motivate students to learn from their mistakes, which ultimately enhances their overall educational experience. Also, the study’s use of multilevel modelling is noteworthy, as it allows the researchers to examine the effects of both within-level and between-level variables on knowledge exploration. This method is instrumental in educational research for exploring the intricate relationship between individual and contextual factors in determining outcome variables (Ruzek et al., 2022). It has also recommended as an effective approach to capture the transient nature of knowledge acquisition process (Murayama et al., 2017; 2022).

Though the study makes some significant contributions, it also has certain limitations. A significant limitation of the study is its ecological validity, as the controlled experimental setting may not accurately reflect real-life learning environments. Another limitation of this study is its generalizability across diverse educational contexts and age groups, as affective and cognitive processes vary significantly with age and educational background. Furthermore, the use of self-reported Academic Motivation Scale (Vallerand et al. 1993), may have introduced social desirability bias. Participants might adjust their responses to appear more socially acceptable, leading to potential overestimations or underestimations of their true motivational levels.

Despite the limitations, the study identifies key factors affecting real-time knowledge exploration among students and opens several avenues for further investigation. One potential direction for future research is the use of activity-based experiments to understand exploratory behaviour comprehensively. While this study focuses only on secondary school students, further research could examine diverse student populations, such as elementary school and university students, to assess the consistency of the empirical findings. Finally, given the significant impact of study variables on knowledge exploration, exploring other cognitive, emotional, and behavioural antecedents may be valuable. This would provide the research community in educational psychology with an in-depth understanding of exploratory behaviour.