The Effect of Growth Mindset Interventions on Students’ Self-Regulated Use of Retrieval Practice

Abstract

Although general growth mindset interventions have been found to improve learning outcomes, relatively little research focused on effects of specific growth mindset interventions. However, domain-specific growth mindset beliefs may be a more accurate predictor of performance within that same domain. The current study investigated the effect of a specific growth mindset intervention designed to enhance self-regulated learning on learners’ adoption of retrieval practice while studying image–name pairs. The impact of this targeted approach was compared to the effects of a broader, general growth mindset intervention, which emphasized the brain’s malleability. Participants were 178 first-year university students, who were randomly assigned to three groups: a general growth mindset group (n = 58), a specific growth mindset group (n = 64), and a control group (n = 56). All groups were informed about the benefits of retrieval practice after the intervention. Our results showed that higher education students can benefit from mindset interventions in terms of their growth mindset beliefs. However, no effects on the use of retrieval practice were found, possibly due to the high difficulty of the learning material. Future research could explore the difficulty of learning materials when investigating the effect of mindset interventions on students’ use retrieval practice during self-regulated learning.

1. Introduction

Numerous studies have explored self-regulated learning (SRL), aiming to identify and examine strategies to empower students to become efficient and effective self-regulated learners (Boekaerts & Corno, 2005; Pintrich, 2000; Schunk & Zimmerman, 1998; Winne & Hadwin, 1998; Zimmerman, 2000). SRL serves as an overarching framework encompassing a myriad of processes essential for learners to effectively manage and direct their learning, a capability crucial for academic success both within and outside educational settings (Boekaerts, 1999; Puustinen & Pulkkinen, 2001; Wong et al., 2019; Zimmerman & Schunk, 2011). In an optimal SRL scenario, learners engage in a cyclical three-phase process to acquire knowledge and achieve their educational objectives (Zimmerman & Moylan, 2009). First, in the forethought phase, learners combine their beliefs about themselves as learners and the learning tasks, to set learning goals, and choose appropriate learning strategies. Then, in the performance phase, learners monitor their learning process during the learning task. Finally, in the self-reflection phase, learners can evaluate their goals, strategies they used and reflect on the possible reasons for their results (e.g., effort and time invested in the learning task, choice of learning strategy, Panadero, 2017; Zimmerman, 2013).

Within the cycle of the three SRL phases, using learning strategies is an important step to help learners improve their learning performance (Ariel & Karpicke, 2018; Dunlosky et al., 2013). Yet, previous research has shown that learners are not always aware of the benefits of using effective learning strategies (Bjork et al., 2013; Kornell & Bjork, 2007). One possible reason is that effective learning techniques such as retrieval practice demand heightened cognitive engagement and effort compared to more passive methods like rote memorization (Sweller et al., 1998). Previous research suggests that a growth mindset enhances resilience to failure and reduces perceived effort during learning (Xu et al., 2021). Therefore, in the current study we investigated the effect of a specific growth mindset intervention focused on self-regulated learning (SRL) and using effective study strategies, on students’ beliefs about SRL, their choices to use retrieval practice and their performance.

1.1. Retrieval Practice

Retrieval practice is a well-studied learning strategy, in which learners self-test the learning materials during their learning process (Karpicke & Blunt, 2011; Roediger & Karpicke, 2006a, 2006b). During self-testing, learners retrieve the information from their memory, and this leads to better retention of learning materials compared to re-studying, especially for long-term retention (Adesope et al., 2017; Agarwal et al., 2012; Karpicke & Roediger, 2008; Roediger & Butler, 2011). This effect of self-testing on retention of learning materials is called the retrieval practice effect or testing effect (Carpenter & DeLosh, 2006; Dirkx et al., 2014; Karpicke, 2017; Rowley & McCrudden, 2020).

Although research has demonstrated the effectiveness of retrieval practice, learners often resist incorporating it into their study routine, instead opting to use it merely as an evaluative tool at the end of their learning process to assess performance (Carpenter, 2023; Hartwig & Dunlosky, 2012; Kornell & Bjork, 2007). Possibly, learners believe that restudying will lead to better performance compared to retrieval practice (Karpicke & Roediger, 2008). For example, Kornell and Son (2009) found that learners judged restudying as more effective, even though they more often chose to test themselves—suggesting that self-testing was used more for monitoring than for promoting learning. Thus, it is crucial to instruct learners on the benefits of using retrieval practice as a learning strategy (Karpicke, 2012; McDaniel et al., 2011).

A multitude of studies have demonstrated that providing participants with instructions on retrieval practice before they commence a learning task can effectively encourage them to incorporate retrieval practice into their learning process (Butler & Roediger, 2007; Karpicke et al., 2009). For example, Broeren et al. (2021) investigated the effect of an online retrieval practice instruction on the self-regulated use of retrieval practice with higher education students in a classroom environment. They provided an online retrieval practice video instruction to the participants in the experimental group, and a neutral video instruction without strategy information to the participants in the control group. In the first and second learning session, participants received the retrieval practice instruction and the learning task (i.e., 20 key concepts of marketing communication in each session). In the third learning session (i.e., transfer session), participants only received the learning task of 20 key concepts without retrieval practice instructions. During the learning sessions, participants studied the concepts in a self-paced way with three choices: (1) re-study, (2) self-test or (3) stop studying. The results indicated that during the first and second study sessions, there were no significant differences in the choices made by participants in the experimental and control groups (i.e., restudy, self-test or stop). However, in the transfer session, participants in the experimental group used the testing option significantly more than participants in the control group. The findings highlight the potential of providing students with explicit instruction and knowledge about effective learning strategies to enhance their self-regulated use of learning strategies.

However, even if students do know about the benefits of effective learning strategies, they still do not always use them during learning (Broeren et al., 2021; Dunlosky et al., 2013; Yan & Schuetze, 2023). Based on the outcomes from study sessions 1 and 2 in the research by Broeren et al. (2021), it appears that instructing students about using retrieval practice does not consistently influence their initial behaviors immediately, such as employing more retrieval practice during the learning process compared to students in the control group. One possible reason for this reluctance to employ retrieval practice could be that effective learning techniques such as active retrieval, spaced repetition, elaboration, and reflection demand heightened cognitive engagement and effort compared to more passive methods like rote memorization (Sweller et al., 1998).

1.2. Growth Mindset

Interestingly, a meta-analysis about students’ mindsets during learning showed that mindsets could influence the use of strategies during SRL (Burnette et al., 2013). In the early research on the implicit theory of intelligence by Dweck (1986), two beliefs about intelligence were proposed. One belief is that individual’s abilities can be continuously improved through efforts (i.e., growth mindset or incremental theory), the other belief is that the level of individual’s ability is fixed, and cannot be changed through hard work (i.e., fixed mindset or entity theory). Numerous studies have shown that mindsets can impact learning performance (Dweck, 2017; Dweck et al., 1995; Dweck & Yeager, 2019; Haimovitz & Dweck, 2016, 2017; Moser et al., 2011; O’Rourke et al., 2014; Paunesku et al., 2015; Yeager et al., 2019).

Studies have demonstrated that promoting a growth mindset can lead learners to adopt distinct achievement goals and learning strategies when compared to maintaining a fixed mindset (Burnette et al., 2013; Dweck & Master, 2012; Xu et al., 2025). That is, individuals with a growth mindset tend to set mastery goals, which involve focusing on developing their own abilities during learning. In contrast, individuals with a fixed mindset tend to set performance goals, which involve aiming to outperform their peers (Dweck, 1986). Additionally, individuals with a growth mindset tend to adopt mastery-oriented strategies, such as investing more time in practicing the skill or knowledge, to stick to their learning goals after experiencing setbacks. In contrast, those with a fixed mindset tend to employ helpless-oriented strategies, such as procrastination, which may lead them further away from their learning goals (Burnette et al., 2013; Dweck & Yeager, 2019; Howell & Buro, 2009). Ultimately, the impacts of growth mindset on the learning process will affect the achievement of their learning goals (Burnette et al., 2013; Dweck & Leggett, 1988; Song et al., 2020).

Researchers have recently proposed the existence of different types of mindsets, specifically distinguishing between domain-general mindsets and domain-specific mindsets (Karlen & Hertel, 2021). Domain-general mindsets encompass beliefs about personal intelligence and overall ability. Domain-specific mindsets could pertain to attitudes towards particular areas such as self-regulated learning, computer science, and mathematics (Karlen & Hertel, 2021). Research has shown that a domain-specific mindset is a better predictor for performance in the same domain and has a stronger relation with the habitual use of learning strategies than the domain-general mindset (Burnette et al., 2020; Hertel & Karlen, 2021; Karlen & Compagnoni, 2017; Karlen et al., 2021). Additionally, individuals can hold different mindsets in different domains. For example, Scott and Ghinea (2013) did a survey with first- and second-year undergraduate students about the difference between a domain-specific and a domain-general mindset. They found that students who have a fixed mindset on computer programming can have a growth mindset on their general intelligence. In addition, the mindset for programming ability was a better predictor for software development practice. Consequently, drawing from these findings, we can infer that learners could hold different mindsets about their general intelligence in contrast to other domain-specific areas or their SRL abilities (e.g., Karlen & Hertel, 2021). Furthermore, a specific SRL growth mindset could potentially be a more accurate predictor of learners’ SRL behaviors compared to a general growth mindset. Hence, it could be promising to support students’ specific SRL growth mindsets to increase the use of effective learning strategies such as retrieval practice.

1.3. Cognitive Load

Together with learners’ mindset and beliefs about retrieval practice during self-regulated learning, the effort required to implement retrieval practice is also a significant factor in the use of it in a self-regulated manner (Storm et al., 2010). Engaging in retrieval practice can be more challenging and necessitate more effort compared to simply re-studying materials (Kornell et al., 2009; Pyc & Rawson, 2009). This perception of retrieval practice as more labor-intensive can hinder learners’ ability to recognize its benefits (F. Paas & van Merriënboer, 2020; McDaniel et al., 2011; Rowland, 2014; Sweller et al., 2019).

Interestingly, Xu et al. (2021) have shown that a growth mindset intervention can lower learners’ perceived cognitive load (i.e., perceived amount of invested mental effort) during learning. They conducted a study with 138 secondary school students to investigate the connection between perceived cognitive load and growth mindset. In the experimental group, participants received a general growth mindset intervention about the malleability of intelligence, while in the control group, participants received an article about basic brain function. Following these interventions, all participants learned about the Doppler effect in a classroom setting. The results showed that participants in the growth mindset group perceived lower cognitive load and had better retention performance than participants in the control group.

By demonstrating to students that they can navigate and regulate their learning environment, the SRL growth mindset intervention may yield more pronounced reductions in perceived cognitive load and improvements in SRL performance compared to general interventions. Furthermore, this reduction in cognitive load could serve as a catalyst for facilitating other demanding learning processes, including the deliberate use of learning strategies such as retrieval practice. Thus, while general growth mindset interventions are valuable, the specific SRL mindset intervention potentially offers a focused approach to empowering students in their learning journey, leading to more significant and enduring benefits for learning.

2. Current Study

Building upon the insights from Xu et al. (2021) and prior research on general and specific growth mindsets (Dweck & Yeager, 2019; Karlen & Hertel, 2021), we expect specific growth mindset interventions to be more promising in supporting students self-regulated learning processes compared to general growth mindset interventions. While general growth mindset interventions promote the belief that abilities can be developed through dedication and perseverance, a specific SRL growth mindset intervention can enhance students’ ability to regulate their own learning processes effectively. Hence, a specific growth mindset intervention could benefit students SRL mindset beliefs, their beliefs about and use of learning strategies such as retrieval practice, decrease their experienced cognitive load, and improve their performance during SRL.

The aim of the current study was to investigate the effects of a specific growth mindset intervention about SRL (SGM), a general growth mindset intervention about individual intelligence (GGM), and a control condition on growth mindset beliefs, retrieval practice beliefs, use of retrieval practice, perceived mental effort, and cued-recall performance of anatomy image–name pairs.

We formulated the following five hypotheses:

Hypothesis 1 (H1): 

Participants who received the SGM intervention were expected to demonstrate higher levels of SRL growth mindset beliefs, but lower levels of general growth mindset beliefs compared to those who received the GGM intervention. Meanwhile, participants in the control condition were expected to exhibit the lowest levels of both SRL and general growth mindset beliefs among the three groups.

Hypothesis 2 (H2): 

Participants who received the SGM intervention were expected to engage more frequently in retrieval practice compared to those in the GGM intervention. In turn, participants in the GGM condition were predicted to utilize retrieval practice more often than those in the control group.

Hypothesis 3 (H3): 

Participants who received the SGM intervention were expected to report lower mental effort compared to those in the GGM condition, who were, in turn, expected to report lower mental effort than participants in the control condition.

Hypothesis 4 (H4): 

Participants who received the SGM intervention were expected to exhibit superior performance in both immediate and delayed cued-recall test compared to those who received the GGM intervention. In turn, participants in the GGM condition were expected to outperform those in the control group on both types of cued-recall tests.

Hypothesis 5 (H5): 

Participants who received the SGM intervention were expected to show higher levels of retrieval practice belief compared to those who received the GGM intervention, who were, in turn expected to show higher levels of retrieval practice belief than participants in the control condition.

3. Materials and Methods

3.1. Participants

Participants were 178 freshmen (n = 58 in GGM condition, n = 64 in SGM condition, n = 56 in Control condition) of the psychology department at a university in the Netherlands. The sample size surpassed the necessary 159 participants determined by an a priori power calculation, which was based on a one-way ANOVA with three conditions, targeting an effect size of Cohen’s f = 0.25, with 80% power and a 5% type 1 error rate.

We checked that none of the participants had received training in growth mindset or retrieval practice and had little prior knowledge of the learning task. The demographic information is provided in

Table 1. Demographic Characteristics of Participants in Three Conditions.

Demographic Characteristics	GGM Condition		SGM Condition		Control Condition
	n	%	n	%	n	%
Gender
Female	47	81.03	52	81.25	46	82.14
Male	10	17.24	11	17.19	8	12.5
Third gender	1	1.72	1	1.5	2	3.57
	M	SD	M	SD	M	SD
Age	19.93	2.67	20.02	1.80	20.09	2.28
Prior-Knowledge a
Anatomy *	1.95	0.61	2.02	0.60	2.32	0.58
Growth mindset	2.64	1.14	2.66	1.04	2.66	1.07
English Level	4.36	0.81	4.13	0.86	4.16	0.65

^a Reflects the knowledge of Anatomy and Growth mindset before intervention. * Reflects a significant difference between conditions.

. There were no significant differences among three conditions in all characteristics except from the Prior-Knowledge of Anatomy. The scales used to measure both Prior-Knowledge and English Level are 5-Likert scales.

A one-way ANOVA revealed a significant effect of condition on prior knowledge of anatomy, F(2, 175) = 6.37, p = 0.002, η_p² = 0.068. Tukey HSD post hoc tests indicated that participants in the control condition reported significantly higher prior knowledge than those in the GGM condition (p = 0.003) and the SGM condition (p = 0.015). No significant difference was found between the GGM and SGM groups (p = 0.807).

3.2. Materials

3.2.1. The General Growth Mindset Intervention

The general growth mindset intervention, adapted from Blackwell et al. (2007) and Xu et al. (2021), focused on the concept of intelligence and incorporated both reading and writing exercises (see Appendix A.1). Participants read an article focusing on the malleability of intelligence and subsequently composed a letter to a fictitious peer struggling with his/her learning of anatomy in medical school and gave them advice to overcome the challenge.

3.2.2. The SRL Growth Mindset Intervention

The newly developed SRL growth mindset intervention was specifically designed to focus on the malleability of the SRL ability, especially for strategy use (see Appendix A.2). It was developed based on the general growth mindset intervention (cf. Blackwell et al., 2007) but instead of the malleability of intelligence as the core of the intervention, SRL abilities were explained using theory on SRL models (Zimmerman, 2013). First, it introduced the theory of SRL and provided the basic knowledge of the three phases of SRL. Then, participants read about the malleability of SRL abilities, for example, “When you learn and practice new learning strategies, these small connections in the brain multiply and become stronger. The more you challenge your brain to practice the new strategies, the more your brain cells grow.”, accompanied by illustrative images depicting neural connections in the brain. After reading the text of the intervention, participants went through the same writing session as the general growth mindset intervention.

3.2.3. The Neutral Intervention

In the control condition, participants received an article, adapted from Xu et al. (2021), which included basic information about the brain function. After reading the article, participants were asked to write a short summary of it (Appendix A.3).

3.2.4. Retrieval Practice Instruction

After receiving one of the three interventions, and before proceeding to the learning task all participants were instructed about the benefits of using retrieval practice as a learning strategy (Figure 1; cf. Ariel & Karpicke, 2018).

3.2.5. Learning Task

Participants were asked to learn 40 anatomical image–name pairs, which were adapted from Hui et al. (2021) (Appendix B). These pairs were divided into 10 units, and each unit included 4 pairs, which were each displayed for 8 s during learning.

To familiarize participants with the learning strategy and task structure, Unit 1 and Unit 2 were used during the learning strategy orientation phase. The actual learning task consisted of Units 3 to 10, which served as the main experimental materials.

3.2.6. Cued-Recall Test

Participants completed each anatomical name; the initial letter was provided as a cue. And the accuracy of the answers was scored as the learning performance.

The 8 experimental units (Units 3–10) each included 4 items. Within each unit, items were assigned to the immediate or delayed test conditions based on a fixed allocation rule: the 1st and 3rd items in each unit were assigned to the immediate test condition, while the 2nd and 4th items were assigned to the delayed test condition.

3.3. Measures

3.3.1. General and SRL Growth Mindset Beliefs

Both the revised Implicit Theories of Intelligence Scale (re-ITIS; De Castella & Byrne, 2015; Appendix C.1) and the revised Implicit Theories of SRL Scale (re-ITSS, cf. Hertel & Karlen, 2021; Appendix C.2), were administered across all three conditions to assess mindsets both pre- and post-interventions. These scales were also employed as a manipulation check to evaluate the impact of the growth mindset interventions. Gain scores for growth mindsets (i.e., Gain Score_GGM = reITIS_post − reITIS_pre; Gain Score_SGM = reITSS_post − reITSS_pre) were analyzed as indicators of intervention effects.

For the re-ITIS, the Cronbach’s alpha of the pre-intervention data was 0.95; the Cronbach’s alpha of the post-intervention data was 0.96.

For the re-ITSS, the Cronbach’s alpha of the pre-intervention data was 0.63, if item 3 was removed, the Cronbach’s alpha was 0.67; the Cronbach’s alpha of the post-intervention data was 0.65, if item 3 was removed, the Cronbach’s alpha was 0.71.

3.3.2. Retrieval Practice Decisions

Participants were asked to answer the multiple-choice question: “Now that you have studied these items twice, do you want to: (A) restudy or (B) self-test?” Learners were free to choose a learning strategy. Based on their choices we calculated the percentage of self-testing choice (cf. Hui et al., 2021).

3.3.3. Mental Effort

A 9-point mental effort rating scale (F. G. Paas, 1992; F. Paas et al., 2003) was used to measure participants’ perceived amount of invested mental effort after they finished each restudy unit and retrieval practice unit, by asking “You studied four image–name pairs. How much mental effort did you invest from very, very little mental effort (1) to very, very much mental effort (9)?”.

3.3.4. Immediate and Delayed Recall Performance

The percentage of correct responses for anatomical names was calculated for both the immediate cued-recall test (completed immediately after the learning task) and the delayed cued-recall test conducted after 7 days (cf. Hui et al., 2021).

3.3.5. Retrieval Practice Beliefs

To determine participants’ retrieval practice beliefs, they were asked to answer the question “How effective is self-testing in helping you to memorize the anatomical image–name pairs?” on a 7-point Likert scale, ranging from (1) extremely ineffective to (7) extremely effective (cf. Hui et al., 2021).

4. Procedure

The experiment was divided into two sessions, separated by 7 days (Figure 2). On day 1, the procedure and learning task were introduced. At the beginning of the experiment, participants were asked to answer some personal details (e.g., name, gender, etc.). After collecting demographics, participants were randomly assigned to one of the three conditions (the GGM, SGM or Control condition). Subsequently, baseline assessments of both general and SRL growth mindset beliefs were conducted. Following this, participants underwent their respective condition’s intervention and then all received identical instructions on retrieval practice. After these activities, the general and SRL growth mindset beliefs were measured once more. Next, we asked participants to study 32 anatomy image–name pairs. During the learning task, the first 2 units (8 pairs) were used to familiarize the participants with retrieval practice and restudy separately. Then, participants were instructed to study 8 units in total, in each unit the 4 pairs were restudied twice and then participants were asked to choose the strategy for the next two trials. In each unit, perceived mental effort was assessed following the final learning trial. After they finished learning of 8 units, participants went through a 3 min distractor arithmetic task to prevent them from rehearsing the materials before the immediate recall test. Then, half of the 8 units participants had studied were used in the immediate recall test. Retrieval practice belief was measured after the immediate recall test. At the end of the first day, the prior knowledge of growth mindset and anatomy among participants was measured. Then after 7 days, the participants went back to the lab to take the delayed recall test which consisted of the remaining half of 8 units. At the beginning of the second session, the general and SRL mindset beliefs were measured again. Also, at the end of the experiment, the retrieval practice belief was measured again.

5. Data Analysis

A one-way Analysis of Variance (ANOVA) was conducted using SPSS Version 27. To complement the null hypothesis significance testing (NHST) framework and ensure analytical consistency across all planned and exploratory comparisons, Bayesian one-way ANOVA (BOA) was also performed in JASP Version 0.17.2, using the default prior for fixed effects (r = 0.5). The Bayesian approach provides a continuous assessment of evidence via Bayes factors, enabling a distinction between lack of significance and actual support for the null hypothesis, and offering a more nuanced interpretation of both significant and non-significant findings (Wagenmakers et al., 2018).

For the one-way ANOVA, partial eta-squared (η_p²) was reported as the effect size measure for ANOVA, with cut-offs for small, medium, and large effects set at 0.01, 0.06, and 0.14, respectively. For the Bayesian analyses, Bayes factors (BF₁₀) were used to evaluate evidence for the alternative model compared to the null, with values between 1–3 indicating anecdotal evidence, 3–10 moderate, 10–30 strong, 30–100 very strong, and values above 100 representing extreme evidence in favor of the alternative model.

Analyses directly testing Hypotheses 1–5 were planned a priori; these included group comparisons on SRL and general growth mindset beliefs, retrieval practice frequency, perceived mental effort, immediate and delayed performance, and retrieval practice beliefs. Additional analyses—i.e., the delayed gain scores for GGM beliefs, delayed gain scores for SGM beliefs, delayed retrieval practice beliefs, and the 7-day change in retrieval practice beliefs—were exploratory and conducted post hoc to further investigate patterns in the data. In addition, to account for the potential influence of pre-existing anatomy knowledge on immediate learning performance, an analysis of covariance (ANCOVA) was conducted with prior knowledge of anatomy as a covariate.

6. Results

The results revealed no significant differences in the baseline scores of GGM beliefs (i.e., reITIS_pre) across the three conditions, F(2, 175) = 0.64, p = 0.53, η_p² = 0.007. Similarly, there were no significant differences in the baseline scores of SGM beliefs (i.e., reITSS_pre) among the three conditions, F(2, 175) = 0.63, p = 0.54, η_p² = 0.007. See

Table 2. Means and Standard Deviations of dependent variables by Condition.

Dependent Variables	GGM Condition		SGM Condition		Control Condition
	M	SD	M	SD	M	SD
Growth mindset baseline
General growth mindset	4.43	0.85	4.24	1.05	4.37	0.89
SRL growth mindset	4.69	0.52	4.68	0.62	4.78	0.44
Gain score
General growth mindset	0.60	0.62	0.30	0.49	0.04	0.33
SRL growth mindset	0.08	0.53	0.26	0.49	−0.01	0.37
Delayed gain score
General growth mindset	0.45	0.67	0.08	0.59	−0.15	0.45
SRL growth mindset	0.06	0.47	0.08	0.58	−0.14	0.37
Retrieval practice decisions	6.40	1.89	6.11	1.86	6.36	1.96
Mental effort	6.62	0.97	6.71	1.17	6.60	1.08
Learning Performance
Immediate cued-recall	0.23	0.13	0.30	0.18	0.27	0.17
Delayed cued-recall	0.09	0.06	0.10	0.08	0.11	0.09
Retrieval practice beliefs
Day 1	4.17	1.56	4.47	1.14	4.14	1.27
Day 8	2.91	1.59	2.92	1.29	2.73	1.37

for an overview of the descriptive data for all dependent variables, which are growth mindset beliefs (Gain score), retrieval practice decisions, mental effort, immediate and delayed learning performance and retrieval practice beliefs.

6.1. Growth Mindset Beliefs

The result of the Gain Score of GGM beliefs showed a significant difference among the three conditions, F(2, 175) = 18.27, p < 0.001, η_p² = 0.17. The Dunnett’s T3 Post hoc results showed that participants in the GGM condition gained more on the GGM beliefs scale compared to participants in the SGM condition (p = 0.01), which in turn gained more than the control group (p = 0.002). The BOA model comparison showed extreme evidence for the condition model over the null model (BF₁₀ = 2.04 × 10⁵), indicating that the observed data were over 200,000 times more likely under the condition model than under the null model. BOA Post Hoc tests provided robust evidence indicating that participants in GGM condition demonstrated a greater increase on the GGM beliefs scale compared to those in the SGM, BF₁₀ = 9.61, and the control condition, BF₁₀ = 4.81 × 10⁵. Furthermore, participants in SGM condition gained more on the GGM beliefs scale compared to those in the control group (BF₁₀ = 27.64).

The One-way ANOVA on the Gain Score of SGM beliefs revealed that there was a significant difference among the three conditions, F(2, 175) = 5.32, p = 0.006, η_p² = 0.06. The Dunnett’s T3 Post hoc showed that the participants in the SGM condition gained more on the SGM beliefs scale than those in the GGM condition (p = 0.03), and the control condition (p = 0.002). However, there was no difference between participants in the GGM group and control group on the SGM beliefs scale. The BOA model comparison showed moderate evidence in favor of the condition model over the null model (BF₁₀ = 5.58), indicating that the observed data were about five times more likely under the condition model than under the null. BOA Post Hoc tests provided substantial evidence indicating that participants in the SGM condition experienced greater improvements on the SGM beliefs scale compared to those in the control condition, BF₁₀ = 31.55. However, there was anecdotal evidence to support that participants in SGM condition gained more on the SGM beliefs scale compared to participants in the GGM condition, BF₁₀ = 1.18, and there was no evidence to support the difference between participants in the GGM and control condition, BF₁₀ = 0.31.

The results of delayed gain scores indicated that for delayed GGM beliefs, there was a significant difference between groups effect, F(2, 175) = 15.81, p < 0.001, η_p² = 0.15. Participants who received the GGM intervention showed the most substantial delayed increase in GGM beliefs scores compared to those who received the SGM intervention (p = 0.004) and the control condition (p < 0.001). Participants in the SGM condition exhibited a higher gain score in GGM beliefs than those in the control condition (p = 0.05). The BOA model comparison showed extreme evidence for the condition model over the null model (BF₁₀ = 2.96 × 10⁴), indicating that the observed data were over 20,000 times more likely under the condition model than under the null model. BOA Post Hoc tests showed that there was strong evidence to support that participants in GGM condition gained more on the GGM beliefs scale compared to those in the SGM condition, BF₁₀ = 21.92, and control condition, BF₁₀ = 7.82 × 10⁴. However, there was insufficient evidence to support that participants in SGM condition gained more on the GGM beliefs scale compared to those in the control condition, BF₁₀ = 2.39.

The results of the delayed SGM gain score showed that there was a significant between groups effect, F(2, 175) = 3.88, p = 0.02, η_p² = 0.04. Participants in the SGM condition (M = 0.08, SD = 0.58) and GGM condition (M = 0.06, SD = 0.47) showed higher delayed gain scores than participants in the control condition (M = −0.1, SD = 0.37). However, there was no significant difference between participants in the SGM and GGM condition. The BOA model comparison showed anecdotal evidence for the condition model over the null model (BF₁₀ = 1.49), indicating that the observed data were about 1.5 times more likely under the condition model than under the null model. BOA Post Hoc tests showed that there was weak evidence to support that participants in SGM condition and GGM condition gained more on the SGM beliefs scale compared to participants in the control condition, BF₁₀ = 3.09; BF₁₀ = 3.66. There was no evidence to support the notion that participants in SGM condition gained more on the SGM beliefs scale compared to participants in the GGM condition, BF₁₀ = 0.20.

6.2. Retrieval Practice

Across all conditions, participants selected retrieval practice in about 6 of the 8 learning units on average, demonstrating consistent use of this strategy throughout the learning task. Analysis of the number of retrieval practice choices indicated that there were no significant differences between conditions, F(2, 175) = 0.41, p = 0.66, η_p² = 0.005. The BOA model comparison showed strong evidence for the null model over the condition model (BF₁₀ = 0.08), indicating that the observed data were about 12 times more likely under the null model than under the condition model. BOA Post Hoc tests showed insufficient evidence that participants in the SGM condition used retrieval practice more frequently than those in the GGM condition (BF₁₀ = 0.27) or control condition (BF₁₀ = 0.25). Similarly, there was insufficient evidence for a difference between the GGM and control conditions (BF₁₀ = 0.20).

No significant differences between condition in retrieval practice beliefs were found, F(2, 175) = 1.13, p = 0.33, η_p² = 0.01. The BOA model comparison showed moderate evidence for the null model over the condition model (BF₁₀ = 0.15), suggesting the data were roughly 6.6 times more likely under the null model. BOA Post Hoc tests indicated insufficient evidence for group differences in retrieval practice beliefs. The SGM condition did not show higher beliefs than the GGM condition (BF₁₀ = 0.37) or control condition (BF₁₀ = 0.52), nor did the GGM condition differ from the control condition (BF₁₀ = 0.20).

No significant differences between conditions in delayed retrieval practice beliefs were found, F(2, 175) = 0.33, p = 0.72, η_p² = 0.004. The BOA model comparison showed strong evidence favoring the null model (BF₁₀ = 0.08), indicating that the observed data were approximately 13 times more likely under the null model. BOA Post Hoc tests indicated insufficient evidence for group differences in delayed retrieval practice beliefs. Specifically, the SGM condition did not show higher beliefs than the GGM condition (BF₁₀ = 0.19) or control condition (BF₁₀ = 0.26), nor did the GGM condition differ from the control condition (BF₁₀ = 0.24) after 7 days. The analysis of the change in retrieval practice beliefs after 7 days revealed a decline in all conditions, which did not significantly differ between the three conditions, F(2, 175) = 0.47, p = 0.62, η_p² = 0.005. The BOA model comparison showed strong evidence favoring the null model over the condition model (BF₁₀ = 0.09), indicating that the observed data were approximately 12 times more likely under the null model than under the condition model. BOA Post Hoc tests indicated insufficient evidence for group differences in the change in retrieval practice beliefs after 7 days. Specifically, changes in the SGM condition did not differ from the GGM condition (BF₁₀ = 0.31) or control condition (BF₁₀ = 0.22), nor did the GGM condition differ from the control condition (BF₁₀ = 0.22).

6.3. Perceived Mental Effort

No significant differences between the conditions in perceived mental effort scores were found, F(2, 175) = 0.17, p = 0.85, η_p² = 0.002. The BOA model comparison showed strong evidence for the null model over the condition model (BF₁₀ = 0.07), with the data being about 15 times more likely under the null model. BOA Post Hoc tests indicated insufficient evidence for group differences in perceived mental effort scores. Specifically, the SGM condition did not show higher score than the GGM condition (BF₁₀ = 0.21) or control condition (BF₁₀ = 0.22), nor did the GGM condition differ from the control condition (BF₁₀ = 0.20).

6.4. Cued-Recall Performance

The immediate learning performance results suggested there is no significant differences between conditions, F(2, 175) = 2.89, p = 0.06, η_p² = 0.032. The Bonferroni Post Hoc analysis indicated that participants in the SGM condition achieved higher accuracy compared to those in the GGM condition (p = 0.05). Yet, as this was not a significant effect, it should be interpreted with caution. There was no significant difference between participants in the GGM condition and control condition (p = 0.22), and no significant difference between participants in the SGM condition and the control condition (p = 0.27). The BOA model comparison provided anecdotal evidence favoring the null model over the condition model (BF₁₀ = 0.69), indicating that the observed data were slightly more likely under the null model than under the condition model. BOA Post Hoc tests showed moderate evidence for higher immediate learning performance in the SGM condition compared to the GGM condition (BF₁₀ = 2.72). However, there was insufficient evidence for performance differences between the SGM and control conditions (BF₁₀ = 0.31), or between the GGM and control conditions (BF₁₀ = 0.44).

To further examine whether pre-existing anatomy knowledge influenced immediate learning performance, an ANCOVA was conducted with prior knowledge of anatomy as a covariate. The analysis revealed a non-significant main effect of condition, F(2, 172) = 0.15, p = 0.86, η_p² = 0.002, and a strong effect of the covariate, F(1, 172) = 21.11, p < 0.001, η_p² = 0.11. The interaction between pre-existing anatomy knowledge and condition was non-significant, F(2, 172) = 0.25, p = 0.78, η_p² = 0.003, indicating that the assumption of homogeneity of regression slopes was met. These results indicate that pre-existing anatomy knowledge substantially predicted immediate performance, and that once this baseline knowledge was statistically controlled, there were no detectable differences between experimental conditions.

The Bonferroni Post Hoc analysis indicated that a marginal higher accuracy in the SGM condition than in the GGM condition (p = 0.065), consistent with the pattern observed in the original ANOVA. Bayesian ANCOVA yielded similar conclusions, with moderate evidence for the model including both condition and the covariate (BF10 = 1702.53) compared to the covariate-only model (BF10 = 1602.94). However, pairwise Bayes factors continued to show only moderate evidence for the SGM > GGM contrast (BF₁₀ = 2.72), and insufficient evidence for other comparisons. These findings reinforce the importance of controlling for prior knowledge, while also suggesting that condition effects should be interpreted with caution.

For delayed learning performance, the results showed that there were no significant differences between the three conditions, F(2, 175) = 1.03, p = 0.36, η_p² = 0.01. In addition, the accuracy in all conditions was no more than 11%, which indicates that the learning materials were quite difficult for participants to remember after a week. The BOA model comparison showed moderate evidence favoring the null model over the condition model (BF₁₀ = 0.14), indicating that the observed data were approximately 7 times more likely under the null model than under the condition model. BOA Post Hoc tests indicated insufficient evidence for group differences in delayed learning performance. Specifically, the SGM condition did not show higher accuracy than the GGM condition (BF₁₀ = 0.29) or control condition (BF₁₀ = 0.22), nor did the GGM condition differ from the control condition (BF₁₀ = 0.52). See

Table 3. Summary of Frequentist and Bayesian One-Way ANOVA Results.

Dependent Variables	F(2, 175)	p	ηp2	BF10 (Model)	Group Differences (Post Hoc)
Gain score
General growth mindset	18.27	<0.001	0.17	2.04 × 105	GGM > SGM (p = 0.012, BF10 = 9.61); SGM > Ctrl (p = 0.002, BF10 = 27.64); GGM > Ctrl (BF10 = 4.81 × 105)
SRL growth mindset	5.32	0.006	0.06	5.58	SGM > GGM (p = 0.03, BF10 = 1.18); SGM > Ctrl (p = 0.002, BF10 = 31.55); GGM ≈ Ctrl (BF10 = 0.31)
Delayed gain score
General growth mindset	15.81	<0.001	0.15	2.96 × 104	GGM > SGM (p = 0.004, BF10 = 21.92); GGM > Ctrl (p < 0.001, BF10 = 7.82 × 104); SGM > Ctrl (p = 0.05, BF10 = 2.39)
SRL growth mindset	3.88	0.02	0.04	1.49	GGM, SGM > Ctrl (BF10 = 3.09; 3.66); GGM ≈ SGM (BF10 = 0.20)
Retrieval practice decisions	0.41	0.66	0.005	0.08	No significant group differences; BF strongly supports null
Retrieval practice beliefs
Day 1	1.13	0.33	0.01	0.15	No group differences; SGM ≈ GGM (BF10 = 0.37); SGM ≈ Ctrl (BF10 = 0.52); GGM ≈ Ctrl (BF10 = 0.20)
Day 8	0.33	0.72	0.004	0.08	No group differences; all BFs < 0.30 favoring null
Gain Score	0.47	0.62	0.005	0.09	No group differences; all BFs < 0.3 favoring null
Mental effort	0.17	0.85	0.002	0.07	No group differences; all BFs < 0.25 favoring null
Learning Performance
Immediate cued-recall	2.89	0.06	0.03	0.69	SGM > GGM (p = 0.05, BF10 = 2.72); others non-significant (BFs < 0.5)
Delayed cued-recall	1.03	0.36	0.01	0.14	No group differences; performance uniformly low; all BFs < 0.6 favoring null

for a summary of the Bayesian one-way ANOVA results across all dependent variables.

7. Discussion

This study investigated the effects of a general and a specific growth mindset intervention combined with a retrieval practice instruction, on the change in general and specific growth mindset beliefs, self-regulated use of retrieval practice, perceived mental effort, retrieval practice beliefs, and the immediate and delayed cued-recall performance when studying word-picture pairs. Our findings confirmed Hypothesis 1, showing that participants who received the SGM intervention gained more on SRL growth mindset beliefs, but less on general growth mindset beliefs, compared to participants who received the GGM intervention or who were in the control condition. Also, participants who received the GGM intervention gained more on general growth mindset beliefs, compared to participants who received the SGM intervention or who were in the control condition. These findings indicate that the general growth mindset intervention had a significant effect on participants’ general growth mindset beliefs, and the SRL growth mindset intervention had a significant effect on participants’ SRL growth mindset beliefs. Our findings align with previous studies involving domain-general and domain-specific growth mindset intervention (Blackwell et al., 2007; Burnette et al., 2020; Xu et al., 2021).

Furthermore, our findings revealed that after 7 days, participants in the GGM condition exhibited a more significant improvement in general growth mindset beliefs compared to those in the SGM condition. However, no notable difference was observed in the gain scores of SRL growth mindset beliefs between the GGM and SGM conditions. Possibly, the general growth mindset intervention has a more enduring effect on enhancing general growth mindset beliefs than the SRL growth mindset intervention has on improving SRL growth mindset beliefs.

In contrast to Hypothesis 2, we found no evidence that participants who received the SGM intervention used retrieval practice more often than those who received the GGM intervention, which in turn were expected to use retrieval practice more often compared to participants in the control condition. The results indicated that participants across all conditions predominantly used retrieval practice during the learning phase, which is consistent with the results discovered by Ariel and Karpicke (2018). A possible explanation for these unexpected findings might be that we provided instructions about the benefits of using retrieval practice in the retrieval practice instruction, and during the learning strategy decision phase, which was after the initial trials of studying the image–name pairs. That is, participants first studied the image–name pairs in the first two trials before choosing to use retrieval practice or restudy, which means they already used the restudy strategy. Participants may have chosen to adopt retrieval practice due to its perceived effectiveness, as emphasized during the study. Alternatively, they may have welcomed the opportunity for a change in their approach to processing the image–name pairs. Future research could adjust the procedure of the experiment to prevent this and clarify under what circumstances participants are more inclined to choose retrieval practice.

Contrary to Hypothesis 3, which posited that growth mindset interventions would reduce perceived mental effort compared to the control group and that the SGM intervention would have a greater effect than the GGM intervention, our findings did not support this. Instead, participants across all conditions reported a moderately high subjective mental effort for the learning task. This consistent level of effort might be attributed to factors identified by Hui et al. (2021) in a similar study: participants’ lack of prior knowledge in human anatomy, which could increase task difficulty, and the brief 8 s time allocation for learning each pair, potentially heightening perceived mental effort. Therefore, future research could manipulate different levels of task difficulty to investigate whether there would be a different assessment of perceived mental effort under different levels of task difficulty.

The results of immediate and delayed learning performance did not confirm Hypothesis 4, which was that participants who received the SGM intervention were expected to outperform those who received the GGM intervention in both immediate and delayed cued-recall tests. Similarly, participants in the GGM condition were expected to outperform those in the control condition on both types of cued-recall tests. Our findings were consistent with those Burnette et al. (2020), who used a domain-specific growth mindset intervention that was focused on the flexibility about one’s computer science ability. Their results showed that even though participants who received growth mindset intervention reported stronger growth mindset beliefs compared to control group, participants in the intervention group had no significant difference on learning grades compared to participants in the control group. From our results, it can be inferred that while learners’ beliefs may change during a single experiment, their learning behaviour, particularly on challenging tasks, may not adapt as swiftly (Macnamara & Burgoyne, 2023). Consequently, learning performance may also lag behind changes in beliefs (Broeren et al., 2021). Given that our study design included only one learning session, participants lacked the opportunity to adjust their learning behaviour, such as their choice to engage in retrieval practice. Therefore, in future research, more than one round of learning should be used to test if there is a delayed effect of growth mindset on changing learning behaviour and performance.

These findings may also be understood in the broader context of the mindset literature. Previous large-scale meta-analyses (Sisk et al., 2018; Macnamara & Burgoyne, 2023) have reported that growth mindset interventions tend to have minimal impacts on academic achievement. Although Burnette et al. (2023) criticized some of the methodological approaches used in earlier meta-analyses and applied multilevel modeling techniques, their study still reported only a modest overall effect size (d = 0.14) for academic outcomes. While the current study does not aim to examine general academic achievement, our results align with this pattern in showing limited behavioral impact of growth mindset interventions within a specific, short-term learning context. As pointed out by Yan and Schuetze (2023), interpreting mindset interventions requires attention to the design features, target outcomes, and timing of measurement. In our case, a single-session design without feedback may not have been sufficient to elicit measurable changes in learning performance. Future research should consider longer-term designs, more ecologically valid tasks, and contextual factors that may modulate intervention effectiveness.

The results of immediate retrieval practice beliefs from session 1 showed that the participants in all conditions did not confirm Hypothesis 5, which expected that a growth mindset can yield a higher belief in the effectiveness of using retrieval practice during learning and the SGM intervention would have a better effect on the retrieval practice belief compared to the GGM intervention. The delayed retrieval practice belief results revealed no significant differences in belief ratings after 7 days among the three conditions. In the current study, the lack of substantial differences in both immediate and delayed learning performance across all conditions likely contributed to the absence of a significant increase in the belief in the efficacy of using retrieval practice across these groups. Even though participants chose to use retrieval practice most of the time, they did not receive feedback about their performance, therefore they may not sense the benefits of using retrieval practice by themselves, which may have resulted in a mediate level of belief. Our findings were consistent with previous research of Hui et al. (2021), who demonstrated that retrieval practice beliefs significantly increased only among participants who perceived benefits from its use. Conversely, those who did not observe an improvement in learning performance after employing retrieval practice did not exhibit a significant change in their beliefs regarding its effectiveness. Accordingly, future investigations could incorporate feedback mechanisms within the learning framework, facilitating the provision of concise and immediate performance evaluations after the application of retrieval practice. This approach stands to refine the assessment of its effectiveness within self-regulated learning contexts.

8. Limitations

There are some limitations to our study. First, the learning materials presented to participants were unfamiliar, and with only 8 s allocated for presentation during learning and tests (as per Hui et al., 2021), the majority of participants reported significant difficulty in recalling the materials following the completion of tests in both sessions 1 and 2. Therefore, we recommend employing tasks of varying difficulty levels in future research to investigate how task difficulty influences the impact of growth mindset interventions on the use of retrieval practice and subsequent performance.

Second, the sample size estimation of the current study was based on an assumed medium effect size (f = 0.25), following Cohen’s (1988) guidelines. This decision was made because, at the time of study planning, not much prior published research was available to provide a more accurate estimate for our specific context. We acknowledge that subsequent studies have reported smaller effect sizes in similar paradigms (Burnette et al., 2023; Macnamara & Burgoyne, 2023; Sisk et al., 2018), suggesting that the current study may have been underpowered to detect such subtle effects. Future research should take these updated findings into account when conducting power analyses.

Third, this study included only one learning session, giving participants no opportunity to adjust their learning strategies in a subsequent round. Also, growth mindset interventions may have a more significant impact on self-regulated learning (SRL) when administered to learners who experience challenges or setbacks in their educational journey (Dweck & Yeager, 2019). Therefore, we recommend using more than one round of learning to test the long-term effects of growth mindset interventions, especially in a challenging learning environment. In addition, although retrieval practice instruction was provided to all participants to ensure equal awareness of the strategy’s benefits, this design choice may have minimized group differences in retrieval practice selection. Future studies may benefit from incorporating a baseline measure of participants’ retrieval practice beliefs and usage to better capture intervention-related changes.

Fourth, we only used retrieval practice choices and learning performance as the indicators of the effect of growth mindset interventions on SRL, while there are many other SRL measurements that could be relevant in relation to growth mindset beliefs (e.g., goal setting, judgment of learning, etc.). Future research could use more SRL indicators to test a broader network between growth mindset intervention and SRL (Yan & Schuetze, 2023).

Additionally, the revised Implicit Theories of SRL Scale (cf. Hertel & Karlen, 2021) used in our study had a relatively low internal consistency (Cronbach’s alpha = 0.63 for the pre-intervention data; 0.65 for the post-intervention data), which may raise concerns about the reliability of observed changes in participants’ SRL growth mindset after intervention. Therefore, future research should consider developing more reliable and stable instruments to assess SRL growth mindset.

However, to further examine the scale’s psychometric properties, we conducted a test–retest analysis between the pre- and post-intervention scores, which yielded a moderate-to-good level of temporal stability (r = 0.62, p < 0.001, 95% CI [0.52, 0.70]). This result suggests that, despite the modest internal consistency, the scale was reasonably stable over time and suitable for capturing change across the intervention period. Still, future research may benefit from employing more rigorous approaches (e.g., longitudinal measurement invariance) and longer follow-up intervals to assess the durability and robustness of intervention effects.

Furthermore, in our study both mental effort and retrieval practice beliefs were measured using single-item scales, which, while commonly used in prior research (e.g., F. G. Paas, 1992; Hui et al., 2021; Xu et al., 2021), may limit the precision and reliability of the constructs. To improve measurement reliability, future research might benefit from including multi-item scales to further examine these constructs.

Finally, the targeted demographic of growth mindset interventions could significantly impact their effectiveness, as indicated by Burnette et al. (2023). Our study, which involved university students, suggests that the impact of growth mindset interventions may be less conspicuous among individuals at the highest level of educational attainment, potentially due to their existing possession of growth mindsets. Consequently, future research endeavors may benefit from comparing the effects of growth mindset interventions on students’ Self-Regulated Learning (SRL) across various educational levels.

9. Conclusions

The current study examined the effect of domain-specific and domain-general growth mindset interventions, coupled with retrieval practice instruction, on growth mindset beliefs, usage of retrieval practice, perceived mental effort, immediate and delayed learning performance, and beliefs about retrieval practice. Our results revealed that a domain-specific intervention targeting SRL abilities effectively enhanced participants’ SRL growth mindset beliefs, while a domain-general intervention focusing on intelligence notably improved general growth mindset beliefs. However, these enhancements in growth mindset beliefs did not affect retrieval practice choices during learning nor performance on immediate and delayed tests. This research is a pioneering effort to combine growth mindset interventions and the self-regulated use of learning strategies. Our findings can provide a basis for future research into the complex interplay between growth mindset and self-regulated learning, providing valuable insights for enhancing SRL and learning outcomes.