Self‐regulated learning profiles including test anxiety linked to stress and performance: A latent profile analysis based across multiple cohorts

Medical educators aim to understand why students differ in performance and stress. While performance and stress are associated with student demographics, school factors and aspects of self‐regulated learning (SRL), it remains unclear how these elements interact within individuals. This multi‐cohort study identified SRL profiles among medical students and explored their associations with performance and stress. Additionally, we examined the identified profiles' associations with gender, migration status and assessment policy.

Discussion: Three distinct SRL student profiles associated with gender, academic performance and perceived stress were identified.Test anxiety had additional value in distinguishing subgroups with differential academic performance and stress.These profiles may aid educators to inform personalised support strategies for novice learners.

| INTRODUCTION
Medical educators have the crucial task of enhancing the learning and academic performance of their diverse student populations while safeguarding their well-being.However, there is still a significant gap in our understanding of the reasons behind variations in the academic performance and mental distress among students.While previous studies have explored associations between student demographics, [1][2][3] school-related factors 1,4,5 and components of self-regulated learning (SRL) 4,6,7 in the context of academic performance and stress, we have yet to understand the distinct patterns of students' SRL behaviours and how its components interact within the context of medical education.

| SRL, well-being and performance
SRL has been defined as 'being metacognitively, motivationally, and behaviourally proactive in the learning process'. 8,9In essence, selfregulated learners are those students who actively plan, set goals, use strategies to execute tasks and monitor their progress and reflect on their performance. 10SRL is usefully assessed using the Motivated Strategies for Learning Questionnaire (MSLQ), as will be detailed in Section 2. 11 This questionnaire consists of two main sections: a 'Motivation' section measuring expectancy, value and affect and a 'Learning Strategies' section, which is further divided into a 'cognitive-metacognitive' section and a 'resource management' section.These components align with the theoretical framework of SRL and are shown in Supporting Information S1 and Figure 1. 10 The connection between SRL and academic performance is well established (Figure 1).4][15] Conversely, test anxiety, which solitarily falls under the domain of affect, is generally associated with lower academic performance. 16,17Additionally, higher levels of value and self-efficacy correlate positively with students' well-being, 18 while test anxiety is associated with increased mental distress and decreased motivation. 19,20Moreover, the use of deep (cognitive) learning strategies such as metacognitive self-regulation, elaboration and organisation and resource management strategies including time management and effort regulation have shown positive associations with superior academic performance (Supporting Information S1). 12,21,22udies involving medical students have demonstrated a positive indirect relationship between deep learning and academic performance through resource management, alongside a negative association between deep learning and academic performance. 8,13is suggests that rather than isolating specific SRL elements, identifying SRL profiles may provide a more comprehensive view of distinct groups of students.In the field of education this person-F I G U R E 1 Visualisation of the conceptual framework applied in this study.The figure shows a simplification of the conceptual framework used by Stegers-Jager et al. 8 The figure illustrates the pathways between value, deep learning, resource management, participation in educational activities and academic performance.The grey labels illustrate the categories of the Pintrich 11,12 framework that are investigated in their paper in relation to academic performance.centred approach, which delves into the creation of subgroups based on the interrelationships among variables, has found several insights. 23,24evious studies in medical students have demonstrated connections between motivational profiles with academic performance, 25 learning features and motivational outcomes 26 and profiles based on learning orientation, behaviour and perceptions of the learning environment have revealed associations with clerkship grades. 27rthermore, students' test anxiety has been speculated to decrease medical students' cognitive strategy use and self-regulation. 11,28Our study builds on this previous work by being the first to create profiles based on self-regulated learning, including test anxiety, and exploring the association between these profiles, student academic performance and perceived stress.Investigating student profiles helps to uncover hidden patterns and relationships that may not be apparent in conventional analysis and allows for the recognition of unique subgroups of students who share common characteristics in their SRL behaviours.Such analyses can give insights in the diversity of learning approaches among students, and variations in academic performance and stress experiences, and could serve as tools for educators and institutions in the future to inform personalised support strategies.

| The role of school and student characteristics
Existing literature predominantly suggests a negative relationship between stress and performance, yet positive associations are also described. 3,4,29,30This intricate relationship is influenced by school, faculty and student factors. 31,32Assessments, for example, can induce stress, 1,4 but can also stimulate performance, with differential effects for male and female students. 1Higher performance standards especially increased the stress levels of female students, whereas it specifically increased the performance of male students. 1is underlines the complex relationship between student performance and stress 33 and the potential role of both student and school factors.
The characteristics of medical schools, including their assessment policies, have been shown to impact students' motivation. 6Particularly, an assessment policy with higher performance standards was found to be associated with higher levels of three SRL components in medical students. 13Furthermore, research indicates that female students generally score higher on motivational constructs. 6,34However, the existing literature is inconclusive about the association between students' migration backgrounds and SRL components. 6,35Taking an intersectional approach, which implies considering how different demographic characteristics reinforce each other, 36,37 our study aims to deepen our understanding of the interplay between student SRL profiles and demographic and school factors within an increasingly diverse student population.

| Current study
Our study investigates how SRL profiles among diverse groups of medical students, encompassing motivational aspects and learning strategies, contribute to our understanding of the complex relationships between student stress and academic performance.We aim to answer this question through a three-step approach as illustrated in (Figure 2).Firstly, we use a person-centred approach to identify profiles among medical students based on both the motivational component F I G U R E 2 Visualisation of the three-step approach to answer the research question.The figure illustrates the research question of the present study in a three-step approach.The measurements that contribute to each element are highlighted in grey.Step 1 highlights the elements (motivation, self-efficacy, test anxiety and learning strategies deep learning and resource management) that contribute to the selection of selfregulated learning profiles in our study.The question marks represent the associations investigated in steps 2 and 3 of our research-question, namely, the association between these identified profiles and student and school characteristics and the association between these profiles and stress perception and academic performance.The grey labels indicate how each component of this research question is measured.
(encompassing value, self-efficacy and test anxiety) and the learning strategies component (encompassing deep learning and resource management) of SRL.
Secondly, we assess how these profiles are associated with assessment policies (in the form of different performance standards) and the (intersection of) student demographics gender and migration background.
Thirdly, we study the association between the identified profiles and the perceived stress and performance in Year 1, aiming to ascertain whether these profiles offer valuable insights the complex relationship between academic performance and stress.

| Context
The present study was conducted at the Erasmus University Medical Center in Rotterdam, the Netherlands.In the Netherlands, the medical school curriculum consists of a 3-year Bachelor's programme (Years 1-3) and a 3-year Master's programme (Years 4-6).Each academic year consists of a total of 60 credits, which can be obtained by students.

| Assessment policy
The assessment policy in this study is represented by the Academic Dismissal policy (AD policy) for Year 1 students.This policy consists of a minimum number of credits students need to obtain to remain enrolled.For cohort 2014 to cohort 2016, the AD policy entailed that students were dismissed from medical school when they did not obtain 100% of Year 1 credits within the first 12 months of their enrolment in medical school (100% AD policy).From cohort 2017 onwards, the assessment policy changed to a 75% AD policy, after an increase in Year 1 repetitions and in help-seeking behaviour by students. 39

| Motivations and strategies for learning
We selected the components of motivations and strategies based on the conceptual framework by Stegers-Jager et al (Figure 1).Their model was based on literature review and subsequently tested and cross-validated in a new independent sample of medical students.
1][42] This 81-item questionnaire is divided into six subscales about motivations and nine subscales about learning strategies.Every item is scored on a 7-point Likert scale (1 = not at all true for me, 7 = very true for me).In the present study, nine out of the total 15 subscales of the MSLQ were used. 8,11,21Test anxiety and self-efficacy were measured through the corresponding MSLQ subscale.The subscales chosen to measure deep learning, resource management and value were selected based on previous studies that have confirmed these subscales cover same latent constructs, 8,13 namely: • Deep learning: mean of subscales elaboration, organisation and metacognition.
• Resource management: mean of subscales time management and effort management.
• Value: mean of subscales task value and intrinsic motivation.
Further information about the used subscales is provided in Supporting Information S1.In the paper by Stegers-Jager, the authors conducted a confirmatory factor analysis (CFA) for these three latent factors measured by MSLQ (deep learning, resource management and value), demonstrating a good fit with the data.

| Student demographics
Students were registered as male or female.In line with the definition of Statistics Netherlands (www.CBS.nl),students were classified as having a migration background when at least one parent was born in a foreign country.

| Perceived stress: cohorts 2014 and 2018
Perceived stress measured with the Dutch validated 14-item PSS-14. 1,38,43The PSS-14 was administered in May of Year 1 for cohorts 2014 and 2018, showing good reliability of, respectively, .888and .866.The questionnaire measures a person's ability to cope with stress and their stress perception during the past month.Items are scored on a 5-point Likert scale, which ranges from 0 (never) to 4 (very often), with a total PSS-14 score between 0 and 56.

| Year 1 performance
Year 1 performance was operationalised as optimal or non-optimal performance in Year 1 (July), indicating whether students obtained all Year 1 credits within the first year or not.Under both assessment policies, compensation was possible for a limited number of courses (max 15 credits) under strict conditions.For cohort 2019, the maximum number of obtainable credits was 58 due to a programme adaptation during the COVID-19 pandemic.As the COVID-19 pandemic started to influence everyday life in March 2020, this only impacted the measure of Year 1 performance for cohort 2019.

| Analysis
Firstly, descriptive statistics of the students' year of enrolment and gender were compared to the overall cohorts.For the MSLQ subscales, reliability was computed, and the scales were composed for the present study.Mean scores on the motivations and learning strategies variables were calculated for the total group and subgroups based on assessment policy and (the combination of) gender and migration background.For subgroups, differences were assessed with t-tests, and effect sizes were reported based on Cohen's d. 44 Secondly, latent profile analysis (LPA) was conducted with standardised values of the five variables representing motivation and learning strategy components (self-efficacy, value, test anxiety, deep learning, resource management).In LPA, the focus is on patterns between different variables, on which profiles of individuals based on similar patterns are constructed. 45LPA has advantages compared to non-latent methods in which clustering is applied, such as K-means clustering. 46One advantage is that in LPA, as opposed to K-means clustering, the probability that a person is a member of a specific cluster is estimated based on the model obtained by LPA. 46,47 LPA, being a member of a specific profile for a specific case is expressed in a membership probability. 45A specific observation (here: student) is assigned to the profile with the highest membership probability. 46Also, in LPA, the Bayesian Information Criterion (BIC) can offer guidance in selecting the optimal number of clusters. 45Given the high exploratory nature of the present study and no prior expectations about the number of profiles, we determined this number based on the BIC, combined with the Integrated Complete-data Likelihoodcriterion (ICL; BIC corrected for entropy), entropy (an estimate of how distinct identified groups are from one another), profiles sizes and cluster probabilities. 47The optimal model selection consists of choosing a model type (Supporting Information S3) and a number of clusters.
Thirdly, links between the identified profiles and assessment policy and student demographics were assessed by chi-squared statistics to determine whether profiles had a different prevalence within different subgroups.Effect sizes of the chi-squared statistics were computed with Cramer's V. 48 Finally, associations between the identified profiles and the variables perceived stress and Year 1 performance were assessed.As a post hoc analysis, these associations between profiles with perceived stress and Year 1 performance were broken down between assessment policies, since previous studies found how links between subgroups with performance and stress can differ between assessment policies. 1,2Differences in mean perceived stress between profiles were assessed with t-tests for which effect sizes were computed with Cohen's d. 44 Associations between profiles and Year 1 performance (optimal versus non-optimal) were assessed with the chi-squared statistic.Effect sizes were computed with Cramer's V. 48 All analyses were performed in RStudio, 49 R version 4.2.1, and R package 'mclust'. 50| RESULTS

| Population versus sample
In total, 1894 students from cohorts 2014 until 2019 completed the MSLQ survey (77% of eligible students).The overall cohorts and responders showed no statistically significant differences, indicating the sample was representative of the complete cohorts in respect to assessment policy, gender and migration background (Table 1).

| Motivations and learning strategies
Computed scales for deep learning, test anxiety, self-efficacy, resource management and value showed sufficient reliability ranging from .70 to .87 (Table 2).More information regarding the construction of the scales can be found in Supporting Information S2.
Under the 100% policy, students showed significantly higher levels of deep learning, test anxiety, self-efficacy and value, although the differences were negligible to small (Table 2).Levels of resource management did not differ between the two assessment policies.Female students showed significantly higher levels of deep learning, test anxiety and resource management and lower levels of self-efficacy (effect sizes small to medium).Value did not differ between male and female students.Finally, compared to students without a migration background, students with a migration background showed significantly lower levels of test anxiety and significantly higher levels of selfefficacy and value.However, these differences were negligible to small.

| Identification of profiles based on motivations and learning strategies
The LPA results show that the most optimal solution based on the BIC is a three-cluster solution (Supporting Information S3).Of the top-five models, the ICL prefers the model with two clusters due to the high overlap between clusters.This suggests a high level of overlap for the three-cluster solution (option 1).This is illustrated by Figure 3, where there is a considerable similarity between cluster 1 and cluster 3 concerning average scores on deep learning, selfefficacy, resource management and value.Since the three-cluster solution is preferred by the BIC, and since these clusters differ on one specific variable (test anxiety), we opted to distinguish these two profiles.We also explored a four-cluster solution, however this resulted in a clear drop in cluster-probability for the added fourth cluster.Therefore, we concluded that a three-cluster solution-as also preferred by the BIC-is the most optimal solution.In Supporting Information S3, the steps taken to explore this solution and alternative solutions are explained in further detail.
The following three profiles were identified with the following overall characteristics (Figure 3): • Profile 1: High test anxiety, high self-regulated learning (TA high SRL high ) Students in this profile show above-average levels of deep learning, self-efficacy, resource management, value and test anxiety.

| Student profiles, assessment policy and student demographics
The profiles were distributed differently under the 100% policy compared to the 75% policy (Table 3).Under the 100% policy, the TA high SRL high profile was significantly more prevalent than under the 75% policy, but with a negligible effect size (X 2 = 8.769, df = 1, p < .01).
The profiles were significantly differently distributed across gender (Table 3).Female students were about 1.5 times more likely to belong to the TA high SRL high profile compared to male students (40.6% versus 26.3%; X 2 = 33.035,df = 1, p < .001,ES = .13,small).The TA low SRL high profile was more prevalent among male students compared to female students (41.8% versus 32.2%, X 2 = 14.863,df = 1, p < .001,ES = .13,small).Students with or without a migration background did not differ significantly in profiles (Table 3).When students' gender and migration background were combined, the largest difference still appeared to be present between female and male students.

| Student profiles, perceived stress and Year 1 performance
We explored the association between the three profiles with students' perceived as well as their Year 1 performance (Table 4).
T A B L E 1 Overview of population and sample.we separately analysed the 100% and 75% policies in a post-hoc analysis, comparable associations were found between profiles and perceived stress (see Supporting Information S4).

Complete cohorts Sample
The Year 1 performance of identified profiles differed from each other with statistical significance (X 2 = 47.465,df= 2, p < .001,ES = .16,small; Table 4).Overall, students in the TA low SRL high profile showed the best Year 1 performance: 82.5% demonstrated optimal Year 1 performance compared to 71.9% (ES = .12,small) and 65.2% (ES = .20,small) for the TA high SRL high profile and the TA moderate SRL low profile, respectively.The Year 1 performance of students in the TA high SRL high profile and the TA moderate SRL low profile differed statistically significantly too, but with a negligible effect.A post hoc analysis broken down to the two different assessment policies revealed that Year 1 performance of the TA high SRL high profile was significantly higher than the Year 1 performance of TA moderate SRL low under the 75% policy (76.3% versus 65.1% optimal performance, X 2 = 8.317,df = 1, p < .01,ES = .12,small) but not under the 100% policy (TA high SRL high versus TA moderate SRL low : 68.4% versus 65.3% optimal performance; see Supporting Information S4).Under both the 75% policy and the 100% policy, students in the TA low SRL high profile outperformed students in the other two profiles.The performance F I G U R E 3 Visualisation of the three identified profiles and their mean scores on motivations and strategies for learning.Mean (SD) scores on self-efficacy, value, test anxiety, deep learning and resource management are shown. 1Prob.= mean membership probability, defined as the probability that a case belongs to the profile they were assigned to, opposed to the other profiles.The T-tests were performed to assess whether the mean scores of students within a profile (represented by the horizontal row in the table /one line in the figure) significantly differed from the mean of the other two profiles combined (represented by the combined remaining two horizontal rows in the table/lines in the figure): gap between students in the TA high SRL high profile and TA low SRL high profile was smaller under the less strict assessment policy, as the effect size of the performance difference decreased from small (100% policy) to negligible (75% policy).

| DISCUSSION
Our study identified three profiles of students that distinguished themselves based on motivation and learning strategy components of SRL, especially test anxiety: TA high SRL high , TA moderate SRL low and TA low SRL high .These profiles, which were identified in the first semester of medical school, were associated with perceived stress in the second semester of medical school and Year 1 performance.On average, students in the TA low SRL high profile showed significantly higher Year 1 performance and lower stress perceived stress compared to students in the other profiles.Differences in performance standards was weakly associated with the identified student profiles based on increased but negligible prevalence of the TA high SRL high profile under the stricter assessment policy.Furthermore, the TA high SRL high profile was 1.5 times more prevalent among female students, whereas the TA low SRL high profile was more prevalent among male students. 51No significant association was found between student migration background and the identified profiles.An intersectional approach combining student gender and migration background did not reveal additional insights.
6][27] Including this affective component in the person-centred approach identified a group of students with high test anxiety within the SRL high group, showing lower performance and experiencing more stress than their SRL high peers.The identification of these profiles, in particular TA high SRL high , may aid in supporting the subgroup of students who are struggling but previously went unnoticed due to their good performance and high levels of SRL.By linking student profiles based on SRL components to both performance and stress, the present study adds to existing literature focusing on student SRL profiles, [25][26][27] thereby further unravelling the complex relationship between stress and performance. 33The identified TA high SRL high and TA moderate SRL low profiles showed, on average, similar levels of perceived stress, with different levels of Year 1 performance, especially under the 75% policy.
The association of lower levels of SRL with lower performance is in line with the SRL theory. 10With similar perceived stress despite different academic performance levels, the question arises whether the origin of perceived stress in the TA high SRL high and TA moderate SRL low profiles was different.However, due to the explorative nature of the current study, speculation on potential underlying mechanisms is premature.A lack of person-environment fit, explained as a mismatch between the students' abilities and skills (low-SRL) and the medical school demands (high-SRL) or learning environment, could be of interest to further explore as possible mechanism causing stress. 52,53ereas the identified profiles did show associations with gender, no or negligible associations were found for student migration background and the medical school assessment policy, respectively.
Notably, students with a migration background had slightly higher levels of self-efficacy, which may be of value to further explore a link between student migration background and components of SRL since existing literature on associations between migration background and components of SRL is mixed and inconclusive. 6,35fferent elements of SRL increased under increased performance standards, which is in line with previous work, 13 but these increases were negligible to small.Under the stricter assessment policy, the proportion of students in the TA moderate SRL low profile remained unchanged.This indicates that a stricter assessment policy does not structurally change students' learning and hence might explain the lack of previously reported long-term performance effects of short-term assessment policy interventions. 2 Nevertheless, the relationships between identified profiles and performance differed among different assessment policies.Especially students in the TA high SRL high profile seemed to benefit with regard to performance from less stringent policies, which is in line with previous studies showing that the association of student subgroups with performance can differ between assessment policies. 1,2e present study has several strengths and limitations.A first strength is the person-centred approach enabling the identification of SRL component combinations within students, including test anxiety, and the link with stress and performance.Inclusion of multiple cohorts controlled for possible cohort effects and resulted in a large sample size.The single-site and explorative nature of the study without expectations about the number of profiles, requires further research to confirm the found profiles.
Our study has limitations to its methodology and study design.
LPA, while useful for exploration, is not oriented towards hypothesis testing or identifying causal mechanisms.LPA group assignments were based on the highest membership probability.While this approach is common, we did not explicitly address the inherent uncertainty in subsequent analyses.
Profiles were based on Year 1 medical students and may change over time.Furthermore, limited perceived stress increased the uncertainty surrounding our identified association between stress and profiles.However, we assume there are no differences between groups with and without PSS-14 measurements and that the measurements are distributed equally across profiles.Despite these methodological constraint, we chose LPA as the most suitable approach to investigate the intricate relationships between SRL, stress and academic performance.
The current findings offer practical implications for medical schools, especially for novice learners entering their transition to firstyear medical students, as included in the present study.Students in the TA low SRL high profile show the desirable performance and stress outcomes.To positively impact performance and well-being of the other students, two discourses can be explored.The most dominant discourse focuses on making the student fit in the academic environment, known as assimilation. 54In this discourse, TA high SRL high students could benefit from psychological guidance or other interventions 55 to cope with test anxiety.As SRL skills can be improved, 56 especially in novices, interventions to improve SRL may support TA moderate SRL low students to perform better. 10The accommodation discourse refers to adjusting the academic environment to adapt to the students, 54 which could support person-environment fit from a different perspective and thereby lower stress levels, 52,53 and increase performance of students.
To conclude, in Year 1 medical students, three different SRL student profiles associated with gender, academic performance and perceived stress can be identified.Test anxiety showed of additional value in distinguishing high-SRL subgroups regarding academic performance and perceived stress.These findings may aid in better supporting or accommodating novice learners 2 Eligible participants of the present study were Year 1 Bachelor students who were enrolled in medical school cohorts 2014-2019, with a total of 2457 students.Year 1 performance and gender data were provided by the university student administration.Data on students' migration backgrounds were obtained through 1 Cijfer HO (1CHO), a national database of students in Dutch higher education.All students who completed the MSLQ were enrolled in the present study.The MSLQ was administered in Year 1 between November and January.In cohorts 2014 and 2018, also the Perceived Stress Scale (PSS-14) 38 was administered in May of the first year.Students who filled out the questionnaires provided written informed consent for the use of their answers in research and for linking their answers to relevant data from student administration.The study was carried out in accordance with the Declaration of Helsinki and was deemed exempt from review after evaluation by the Medical Ethics Committee of Erasmus MC Rotterdam (MEC-2014-387 and MEC-2019-0448).

• Profile 2 :
Moderate test anxiety, low self-regulated learning (TA moderate SRL low ) Students in this profile show below-average levels of deep learning, self-efficacy, resource management and value.They show average levels of test anxiety.• Profile 3: Low test anxiety, high self-regulated learning (TA low SRL high ) Students in this profile show above-average levels of deep learning, self-efficacy, resource management and value.They show below-average levels of test anxiety.
Mean and standard deviation for motivations and learning strategies for each subgroup and the total sample.
a Reliability for two subscales based on Spearman-Brown.b Reliability was based on three subscales (Cronbach's alpha).
Link of profiles with Year 1 performance and perceived stress.