research title about learning strategies

The impact of learning strategies on the academic achievement of university students in Saudi Arabia

Learning and Teaching in Higher Education: Gulf Perspectives

ISSN : 2077-5504

Article publication date: 11 February 2022

Issue publication date: 22 February 2022

This study aimed to investigate the learning strategies adopted by Saudi university students and explore the differences in the use of learning strategies due to gender and academic achievement.

Design/methodology/approach

The study utilized a cross-sectional descriptive analytic approach and adopted the brief “ACRA-C” learning strategies scale. The study sample consisted of 365 students enrolled at a Saudi university selected using the random clustering technique.

The study revealed that microstrategies and study habits are the most preferred strategies by Saudi university students. Statistically significant differences in the use of learning strategies were found between male and female students in favor of the female students. The study also found that learning strategies are a significant predictor of students' academic achievement.

Research limitations/implications

The study was limited to one college in one Saudi university. Future studies should use larger samples from different colleges and universities in Saudi Arabia and incorporate a variety of measures of academic achievement, such as students' grades in specific courses rather than the overall grade average.

Originality/value

While there are a number of studies that investigated the use of learning strategies by students, there is a lack of such research in the higher education context of Saudi Arabia. Hence, the current study contributes to closing this gap in the literature by looking at the use of learning strategies by university students in Saudi Arabia and the relationship between strategy use, gender and academic achievement.

• Learning strategies
• Saudi higher education
• Academic achievement

Almoslamani, Y. (2022), "The impact of learning strategies on the academic achievement of university students in Saudi Arabia", Learning and Teaching in Higher Education: Gulf Perspectives , Vol. 18 No. 1, pp. 4-18. https://doi.org/10.1108/LTHE-08-2020-0025

Emerald Publishing Limited

Copyright © 2021, Yousef Almoslamani

Published in Learning and Teaching in Higher Education: Gulf Perspectives . Published by Emerald Publishing Limited. This article is published under the Creative Commons Attribution (CC BY 4.0) licence. Anyone may reproduce, distribute, translate and create derivative works of this article (for both commercial and non-commercial purposes), subject to full attribution to the original publication and authors. The full terms of this licence may be seen at http://creativecommons.org/licences/by/4.0/legalcode .

Introduction

Traditional rote-learning memorization has been the dominant learning strategy by students in educational institutions in the Kingdom of Saudi Arabia (KSA). This emphasis on rote memorization is responsible to a great degree for Saudi students being passive recipients of information in the classroom ( Al-Seghayer, 2021 ; Pordanjani & Guntur, 2019 ; Kim & Alghamdi, 2019 ).

Recently, in KSA, there has been substantial interest in raising students' awareness of learning strategies in an effort to increase the quality of learning in educational institutions and satisfy preestablished global performance standards, such as the KSA national accreditation requirements established by the National Commission of Academic Accreditation and Assessment (NCAAA). The accreditation certificate is a significant indicator of educational quality, and it assesses four aspects of the educational system: curriculum, instructors, teaching strategies and students. In terms of the student indicators, performance is the first measurement of learning quality ( Vermunt & Vermunt, 2017 ), while learning is measured through attainment or accumulative achievements, such as exam results. Ali, Medhekar and Rattanawiboonsom (2017) argued that student achievement in a higher education institution can be improved through several critical factors namely, the quality of the staff, the inclusion of information technology and appropriate learning strategies. Thus, a number of local studies have investigated the role and impact of instructors in promoting student achievement and learning. For example, Bashir, Lockheed, Ninan and Tan (2018) asserted that pedagogical practice and instructor knowledge play a critical role in increasing student learning. Similarly, Buchori, Setyosari, Dasna, Degeng and Sa'dijah (2017) established that instructors' strategies and techniques determine students' roles, activities and achievement in the learning process and likewise foster students' responsibility for their learning. Other studies investigated learning strategies which can help students acquire information and take an active role in the learning process (e.g. McMullen, 2009 ; Shehzad, Razzaq, Dahri, & Shah, 2019 ).

Research on learning strategies has shown that students may adopt more than one learning strategy since the different academic tasks and their nature require different processing strategies, which range from simple to more complex strategies. Some studies established that the learning strategies could be a good predictor of academic achievement (e.g. Pennequin, Sorel, Nanty, & Fontaine, 2010 ; Muelas & Navarro, 2015 ; Pinto, Bigozzi, Vettori, & Vezzani, 2018 ; Tan, 2019 ), while others found that the relationship between learning strategies and academic achievement was negative such as in Vettori, Vezzani, Bigozzi and Pinto (2020) . Furthermore, a few studies did not find any association between learning strategies and student performance (see Tariq et al. , 2016 ). In their study, Chiu, Chow and Mcbride-Chang (2007) found that different contextual factors such as the economic and cultural background of the students may substantially affect the association between learning strategies and academic achievement.

Despite the extended research conducted investigating the relationship between the use of learning strategies and student academic performance, there is lack of evidence on the use of learning strategies by Saudi students. Therefore, this study explores the learning strategies adopted by Saudi university students in the education process in light of the country's efforts to raise the quality of teaching and learning in its educational institutions.

Literature review

Learning strategies are defined as a set of approaches that learners use to acquire information and knowledge, such as taking notes, organizing information, summarizing and coding ( Muelas & Navarro, 2015 ). There is a difference between learning style and learning strategies. Learning style is used to describe the information processing routines associated with students' personalities, whereas learning strategies refer to students' learning approaches in specific learning activities and learning situations ( Curry, 1990 ; Li, Medwell, Wray, Wang, & Xiaojing, 2016 ).

Effective learning strategies refer to techniques and approaches learners use to achieve the acquisition, storage, retention, recall and adoption of knowledge. Cognitive learning theories consider learners as primary participants in the education process in which their role goes beyond passively acquiring information to being active participants. Consequently, students not only receive information and knowledge but also perform mental activities to process and adopt information effectively ( Shi, 2017 ). Accordingly, learners have a wide range of sources and are free to select their learning strategies, direct their learning process and control their tendencies and emotions to serve their learning objectives ( Díaz, Zapata, Diaz, Arroyo, & Fuentes, 2019 ).

Determine the information that is most significant by extracting keywords, ideas and models.

Make notes that are more frequently used within classroom time, which help students to recall the information mentioned by the lecturer.

Retrieve relevant information associated with the constructivist learning approach, which relies on making associations among prior information and newly acquired information.

Organize the content and material using the specific plan and obvious objectives previously formulated by learners.

Elaborate on the content of the material and course sources, extract conclusions and extrapolate the information.

Summarize the information into general ideas and concepts and determine the more important relationships and conceptual definitions.

Monitor their memorization and comprehension periodically to ensure their understanding and their knowledge.

Similarly, the study of Montero and Arizmendiarrieta (2017) explicated 10 learning strategies consisting of elaboration, time and effort, perseverance, organization, classmates' support, metacognition, self-questioning, the study environment, repetition and instructors' help. Furthermore, Juste and López (2010) identified seven learning strategies that include the planning and reinforcement of self-esteem, classification, problem-solving, repetition, cooperation, deduction and inference, and prediction and assessment. Apart from identifying specific strategies, Muelas and Navarro (2015) classified strategies into four main categories (i.e. information acquisition strategies, information coding strategies, information retrieval strategies and processing support strategies), while Vega-Hernández, Patino-Alonso, Cabello, Galindo-Villardón and Fernández-Berrocal (2017) identified three main categories of learning strategies: cognitive and learning control strategies, learning support strategies and study habits.

Further studies have attempted the classification of learning strategies into micro and macrostrategies ( Jiménez, García, López-Cepero, & Saavedr, 2017 ). Planning and self-regulation are the main pillars of macrostrategies while summarizing and highlighting information are related to tasks and situations that are present in microstrategies. According to Nikou and Economides (2019) , homework is one of the main examples of a microlearning strategy, and this explains why microstrategies are often used among students. Microlearning delivers learning through small and short units within short, focused activities. In microlearning, students summarize and highlight content to obtain smaller units, such as definitions, formulas and brief paragraphs. Conversely, the concept of macrostrategies is seen as a set of approaches encompassing monitoring, revising, checking and self-assessment. Macrostrategies are more general and developmental, and they cannot be directly defined.

Another classification associated with the use of learning strategies was proposed by Rosário et al. (2015) who stated that students have to be self-regulated to control their learning and effectively implement learning strategies. Therefore, students must acquire three types of knowledge: declarative, procedural and conditional knowledge. Declarative knowledge includes information about various learning strategies. Procedural knowledge includes knowing the appropriate way to apply the different learning strategies. Finally, conditional knowledge identifies the proper context to implement a specific learning strategy.

In addition to identifying and classifying the different learning strategies that students employ, a number of studies were carried out to examine the different preferences among students when adopting learning strategies. Vega-Hernández et al. (2017) explored the differences in learning strategy utilization among students according to gender and age and found that male students preferred learning support strategies and study habits, while female students used cognitive and learning control strategies more frequently. Díaz et al. (2019) also revealed that studying in a group, learning through graphic expression and focusing on information synthesis are most commonly used by university students. In a recent study, Tan (2019) found that students rarely used surface or strategic learning strategies, while they frequently used deep learning strategies, but at a moderate level, thus exhibiting less interest in reading and solving word and numeric problems in math.

The subject area has also been found to have an effect on the use of learning strategies. For example, Muelas and Navarro (2015) investigated student strategy use in three main subject areas: language, math and social sciences. In the language subject, the information coding and information recovery strategies were found to be the most significantly related to higher achievement. The coding strategy was the only strategy that had a significant correlation with higher achievement in math and social science subjects. Muelas and Navarro (2015) argued that teaching learning strategies can be a remedial solution for low student achievement, and they illustrated how to exploit brain competencies through learning strategies to improve academic achievement.

Apart from academic achievement, studies have also looked at other psychological aspects in the context of effective use of learning strategies. For example, Tan (2019) concluded that the use of learning strategies has a moderating effect on the relationship between self-concept and problem-solving skills in students studying mathematics. Similarly, Montero and Arizmendiarrieta (2017) found that remedial interventions in enhancing the use of learning strategies improved student motivation and learning beliefs. Vega-Hernández et al. (2017) also found the use of learning strategies had a positive relationship with perceived emotional intelligence (repair, attention and clarity).

While there are a number of studies that investigated different aspects of the use of learning strategies by university students, there is a lack of such research in the higher education context of Saudi Arabia. Hence, the current study contributes to closing this gap in the literature by looking at the use of learning strategies by Saudi university students and the relationship between strategy use and academic achievement. The research question that guided the present study was: “What is the impact of learning strategies on the academic achievement of Saudi university students?” The study further explored whether gender makes any difference in the selection and use of learning strategies.

Methodology

The study adopted a cross-sectional descriptive analytic approach and applied a quantitative method using a scale as a data collection tool. The study intended to examine the adopted learning strategies among students regardless of whether they had a good basic knowledge of learning strategies (i.e. used the learning strategies intentionally or not).

Participants

The study population comprised all students enrolled in the College of Education at a university in Saudi Arabia. First, the participants of the study were selected using the clustering technique. Four degree programs were identified: Diploma, Bachelor, Master and Doctorate. Then, the participants from each degree program were selected using the stratified random technique to include a variety of the population in the sample. The study selected students enrolled in the College of Education to avoid differences in the use of learning strategies due to the subject area. Thus, the target population consisted of 2,870 female students and 999 male students according to the admission and registration department of the university. The study sample consisted of 365 students, which means that the results can be generalized to all students enrolling in the College of Education at the target university (see Krejcie & Morgan, 1970 ). Table 1 shows that the gender distribution of the sample was balanced (49% female and 51% male). The majority of the participants were enrolled in a bachelor's degree program (81.9%). Participants' grade point average (GPA) varied: 44.9% had very good grades, 34.5% had good grades, 18.9% had excellent grades and 1.6% had passing grades. Participants were mainly in their final year (54.8%) and third year (25%).

Data collection instrument

The study adopted the higher education version of the brief “ACRA-C” learning strategies scale by Jiménez et al. (2017) (see Appendix 1 ). The scale assesses the strategies used by students during the learning process in the university. The original ACRA-C scale was adapted to the study context and the scale used in the study comprised 22 items (17 items for learning strategies and 5 items for learning habits). Participants were asked to evaluate each item using a four-point Likert scale according to the knowledge process (from 1 = Never use to 4 = Always use). The knowledge process is anchored mainly on the following strategies: cognitive and learning control strategies, learning support strategies and study habits. The 22 items were further organized into four main categories: microstrategies (Items 1–5), keys of memory and metacognition (Items 6–10), emotional-social support (Items 11–17) and study habits (Items 18–22). Microstrategies are strategies that control leaning (e.g. “I make summaries after underlining”). Keys of memory and metacognition referred to the ability to self-regulate the learning process (e.g. “It helps me if I recall events or anecdotes to remember”). Emotional-social support referred to the personal motivational aspects and learning support from surroundings (e.g. “I study hard to feel proud of myself”). Study habits referred to what students do habitually (e.g. “I try to express what I have learned in my own words, instead of repeating literally what the teacher or the book says”). A sociodemographic section was added to the scale. This section recorded various types of information about the participants such as their degree, gender, college enrollment, GPA and years of study.

The instrument was translated into Arabic prior to distribution to the sample. In order to ensure that the respondents understood the questions, the instrument was presented to a panel of academics in the field to ensure the translated scale was linguistically and culturally valid. Also, the scale was presented to five students who were from the study population but were not included in the study sample to ensure that they comprehended the items fully. Furthermore, the reliability and validity of the scale were measured. The reliability was measured using a split half (Guttman coefficient = 0.657) and Cronbach's alpha for each dimension and the total scale ranged from 0.658 to 0.777, representing an acceptable level of internal consistency (see Table 2 ). Furthermore, the total score of the instrument was 0.726, indicating good consistency.

To test the validity of the instrument, exploratory factor analysis (EFA) was conducted. According to the Kaiser–Meyer–Olkin (KMO) test, the sample was adequate to run the EFA test (KMO = 0.707; Bartlett's sphericity p  = 0.000). The results found that the variance (eigenvalues) of the instrument's items ranged from 1 to 3.39, and the commonalities of all items were higher than 0.4. The results showed that four factors can be retained by eliminating items that are not saturated by any factor (<0.4), as shown in Table 3 . The instrument is divided into four main dimensions: microstrategies, keys of memory and metacognition, emotional support and study habits. The EFA results are similar to the results obtained by Jiménez et al. (2017) . Therefore, the factors were named the same as those in Jiménez et al. (2017) : microstrategies, keys of memory and metacognitive strategies, social-emotional supports and study habits.

Data analysis

The variance of the learning strategies among participants due to gender and GPA was investigated using covariance tests such as the t -test. Then, the combination of bivariate correlation and regression tests was used to investigate the impact of learning strategies on the students' performance.

The central tendency and dispersion of participants' responses were measured for each dimension, as shown in Table 4 . Participants reported frequent use of all learning strategies in their learning and a preference for microstrategies and study habits compared to the rest of the learning strategies. The kurtosis values for all dimensions excluding “study habits” were positive, which show peaked distributions, while “study habits” showed a flatter distribution.

Furthermore, to investigate the differences in the participants' responses due to gender, the t -test was used, and the results are shown in Table 5 . The female participants reported a significantly higher level of use overall ( M  = 3.24; t (363) = 5.689, p  = 0.000) and also for each category of strategies: microstrategies ( M  = 3.28, SD = 0.504; t (363) = 3.79, p  = 0.000), keys of memory and metacognition ( M  = 3.26; t (363) = 4.65, p  = 0.000), emotional and social support ( M  = 3.21; t (363) = 3.75, p  = 0.000), study habits ( M  = 3.24; t (363) = 3.75, p  = 0.000), when compared to the male participants.

Furthermore, the study investigated the differences in the use of learning strategies using academic achievement and gender as the predictors. The results are shown in Table 6 . There was no difference in the learning strategies among students who achieved “passing” grades. However, in students with “good,” “very good” or “excellent” grades, there were significant differences found in the use of learning strategies in favor of the female students.

According to Table 6 , female students who achieved “very good” grades showed higher overall use of learning strategies than males with the exception of “emotional-social support.” However, females who achieved “excellent” grades surpassed the males even in “emotional-social support” along with “study habits” and the overall use of learning strategies, while there was no difference between the genders in “microstrategies” and “keys of memory and metacognition” in this GPA group.

Table 7 shows the results of the linear regression test seeking to discover the impact of learning strategies on student achievement. According to the results, there is a positive relationship between the use of learning strategies and student achievement, where learning strategies can explain 8% of the variance in student achievement. In addition, the learning strategies were statistically significant in predicting student achievement ( F (1, 363) = 34.816, p  < 0.05).

Moreover, a multiple regression test was conducted to investigate the source of the impact of various learning strategies on students' achievement. To conduct a multiple linear regression, multicollinearity has to be checked first. In this study, all variance inflation factors (VIFs) were less than 3, which means that there was no multicollinearity between the learning strategy dimensions, while linearity between the learning strategy dimensions and students' achievement was diagnosed. Another assumption that had to be examined before conducting a multiple linear regression was the normality of the residuals using the Q-Q plot, as shown in Figure 1 in which all data points are so close to the diagonal line; thus, they are normally distributed.

As can be seen in Table 8 , the overall model (microstrategies, keys of memory and metacognition, emotional-social support and study habits) was a significant predictor of student achievement ( F (4, 360) = 10.167, p  < 0.01), where the model explained 10% of the variance in academic achievement and had an appositive mild correlation ( R  = 0.31). The significant contributors of the model were microstrategies ( β  =  0.138, p  = 0.013 < 0.05) and keys of memory and metacognition ( β  =  0.196, p  = 0.001 < 0.01). These two categories were the main sources of the effects on student achievement.

The present study utilized a scale to examine Saudi students' use of learning strategies and the extent to which strategy use is related to academic achievement and gender. The results presented a high preference for microstrategies by students. This can be explained by the fact that in Saudi universities, students are encouraged to use microstrategies like summarizing and highlighting information rather than macrostrategies such as self-regulated learning and planning for learning (see Alhaisoni, 2012 ; Al-Otaibi, 2004 ). In the majority of the lectures delivered in Saudi universities, students are only passive recipients of information, summarizing and highlighting what the instructor disclosed during the lecture, using a specific textbook for reference ( Al-Seghayer, 2021 ; Pordanjani & Guntur, 2019 ; Kim & Alghamdi, 2019 ). This contradicts the results for university students in Lima in Díaz et al . (2019) where students preferred metacognitive strategies and information processing strategies. Study habits which ranked second in this study explained the high level of self-regulation that Saudi students have to control their learning, and this is aligned with the higher education norms in Saudi Arabia, which use mostly a student-centered curriculum. Therefore, students have to assume responsibility for their learning. Accordingly, students always seek summaries and short focus activities to help them acquire information. Nevertheless, the descriptive data also referred to a lack of emotional-social support to students. This could be attributed to the poor educational content, which does not meet students' interests or their educational needs ( Alenezi, 2020 ; Khan, 2019 ).

The results of the study further revealed differences in the frequency of using the various learning strategies, and the overall academic achievement, with female Saudi students showing a higher use of learning strategies. Previous studies in other parts of the world have also shown that female students have a higher level of competence and willingness to perform better in their academic programs ( DiPrete & Buchmann, 2013 ; Tariq et al. , 2016 ; Quadlin, 2018 ). This result is also in agreement with the results obtained by Vega-Hernández et al. (2017) . Furthermore, female students with “good,” “very good” or “excellent” grades showed significant differences in their use of learning strategies compared to male students. However, this was not the case when comparing male and female students with low grade achievement. This makes sense since these students are not successful learners and they therefore do not use learning strategies that much regardless of their gender. In the case of the highest GPA students, there was no difference in all learning strategies except in the emotional-social support category with female students outperforming the male students. These students are highly motivated and competitive with females being extra determined to prove themselves in a patriarchal and male dominated society making the emotional-social support strategies all the more important. These results taken together show that learning strategies have a significant effect on students' academic achievement and they have clear implications for faculty in Saudi universities who have to use numerous and various teaching strategies to induce students' use of appropriate learning strategies especially among the weaker students. Ali et al . (2017) reported that both the quality of the staff and appropriate teaching and learning methods are factors that affect student learning at university. The findings of the current study contribute valuable insight into how faculty in Saudi universities may help develop students' use of appropriate learning strategies.

Finding differences in the use of learning strategies between male and female students of varying GPA levels encourages further investigation of the association between learning strategies use and students' academic performance. In this study, learning strategies explained 8% of the variance in student achievement. The microstrategies and keys of memory and metacognition were the main sources of the effects on student achievement, which means that only these two main strategies statistically significantly predicted the achievement. In addition, the overall model used in this study (microstrategies, keys of memory and metacognition, emotional-social support and study habits) was a significant predictor of student achievement, in which the model explained 10% of the variance in academic achievement. This is in agreement with other empirical studies that support the positive relationship between the use of learning strategies and academic achievement ( Pennequin et al. , 2010 ; Pinto et al. , 2018 ). Furthermore, the evidence presented in this study contradicts studies that refuted any association between learning strategies and student achievement or performance (such as Tariq et al. , 2016 ).

Succinctly, the results revealed that there is a positive relationship between learning strategies and student achievement with the frequency of use of learning strategies significantly predicting the academic achievement of students. Furthermore, Saudi female students were found more eager to use learning strategies than male students, especially in higher GPA levels.

The study assessed the impact of Saudi university students' use of learning strategies on their academic achievement. The study adopted the higher education version of the brief “ACRA-C” learning strategies developed by Jiménez et al. (2017) and divided learning strategies into four main categories: microstrategies, keys of memory and metacognition, emotional-social support and study habits. A total of 365 female and male university students at a College of Education participated in the study. Results showed statistically significant differences in the use of learning strategies due to gender in favor of the female students, which implies that male students have to improve their use of learning strategies and study habits. The study also found that the use of learning strategies significantly predicted student achievement, particularly the microstrategies and keys of memory and metacognition. This implies that students have to pay more attention to the use of these learning strategies if they are to enhance their academic performance.

Based on the study results, it is recommended that training programs on learning strategies be introduced to enrich Saudi students' knowledge and utilization of learning strategies. Also, the training program has to consider the students' gender and their academic level. Furthermore, students have to grasp the significance of the learning strategies as a facilitating tool to increase their academic achievement.

While the study made a valuable contribution, it was limited to one college in one Saudi university. Future studies should use larger samples from different colleges and universities in Saudi Arabia and incorporate a variety of measures of academic achievement, such as students' grades in specific courses rather than the overall grade average.

Despite its limitations, the current study contributed to the field of learning strategy use and filled a gap in the literature by shedding light on the Saudi Arabian context. By examining the relationship between strategy use, academic achievement and gender, it makes an important contribution to Saudi higher education and provides a map to help improve the quality of higher education and student achievement in university.

Normal Q-Q plot of the standardized residual of the regression (DV: student achievement)

Demographic characteristics of the participants ( N  = 365)

Reliability of the scale

Exploratory factor analysis of the instrument (four factors)

The results of the mean comparison t -test according to gender ( N  = 365)

Al-Otaibi , G. N. ( 2004 ). Language Learning Strategy use Among Saudi EFL Students and its Relationship to Language Proficiency Level, Gender and Motivation . Dissertation . Indiana University of Pennsylvania .

Al-Seghayer , K. ( 2021 ). Characteristics of Saudi EFL learners' learning styles . English Language Teaching , 14 ( 7 ), 82 – 94 .

Alenezi , A. M. ( 2020 ). The relationship of students' emotional intelligence and the level of their readiness for online education: A contextual study on the example of university training in Saudi Arabia . The Education and Science Journal , 22 ( 4 ), 89 – 109 . doi: 10.17853/1994-5639-2020-4-89-109 .

Alhaisoni , E. ( 2012 ). Language learning strategy use of Saudi EFL students in an intensive English learning . Asian Social Science , 8 ( 13 ), 115 – 127 . doi: 10.5539/ass.v8n13p115 .

Ali , M. M. , Medhekar , A. , & Rattanawiboonsom , V. ( 2017 ). Quality enhancement in teaching-learning strategies of Bangladesh: A qualitative assessment . International Journal of Advanced Trends in Technology: Management and Applied Science (IJATTMAS) , 3 ( 1 ), 121 – 147 .

Bashir , S. , Lockheed , M. , Ninan , E. , & Tan , J.-P. ( 2018 ). Facing forward: Schooling for learning in Africa . The World Bank . Available at: https://www.worldbank.org/en/region/afr/publication/facing-forward-schooling-for-learning-in-africa .

Buchori , A. , Setyosari , P. , Dasna , W. , Degeng , N. S. , & Sa'dijah , C. ( 2017 ). Effectiveness of direct instruction learning strategy assisted by mobile augmented reality and achievement motivation on students cognitive learning results . Asian Social Science , 13 ( 9 ), 137 – 145 .

Chiu , M. M. , Chow , B. W.-Y. , & Mcbride-Chang , C. ( 2007 ). Universals and specifics in learning strategies: Explaining adolescent mathematics, science, and reading achievement across 34 countries . Learning and Individual Differences , 17 ( 4 ), 344 – 365 . doi: 10.1016/j.lindif.2007.03.007 .

Curry , L. ( 1990 ). A critique of the research on learning styles . Educational Leadership , 48 ( 2 ), 50 – 56 .

Díaz , M. A. , Zapata , N. A. , Diaz , H. H. , Arroyo , J. A. , & Fuentes , A. R. ( 2019 ). Use of learning strategies in the university. A case study . Propósitos y Representaciones Monographic: Advances on Qualitative Research in Education , 7 ( 1 ), 10 – 32 .

DiPrete , T. A. , & Buchmann , C. ( 2013 ). The Rise of Women: The Growing Gender Gap in Education and What it Means for American Schools . NY : Russell Sage .

Jiménez , L. , García , A.-J. , López-Cepero , J. , & Saavedr , F.-J. ( 2017 ). The brief-ACRA scale on learning strategies for university students . Revista de Psicodidáctica , 23 ( 1 ), 63 – 69 . doi: 10.1016/j.psicod.2017.03.001 .

Juste , M. P. , & López , B. R. ( 2010 ). Learning strategies in higher education . International Journal of Learning , 17 ( 1 ), 259 – 274 . doi: 10.18848/1447-9494/CGP/v17i01/46813 .

Khan , S. ( 2019 ). A comparative analysis of emotional intelligence and intelligence quotient among Saudi business students' toward academic performance . International Journal of Engineering , 11 , 1 – 10 . doi: 10.1177/1847979019880665 .

Kim , S. Y. , & Alghamdi , A. K. ( 2019 ). Female secondary students' and their teachers' perceptions of science learning environments within the context of science education reform in Saudi Arabia . International Journal of Science and Mathematics Education , 17 , 1475 – 1496 . doi: 10.1007/s10763-018-09946-z .

Krejcie , R. V. , & Morgan , D. W. ( 1970 ). Determining sample size for research activities . Educational and Psychological Measurement , 30 ( 3 ), 607 – 610 . doi: 10.1177/001316447003000308 .

Li , Y. , Medwell , J. , Wray , D. , Wang , L. , & Xiaojing , L. ( 2016 ). Learning styles: A review of validity and usefulness . Journal of Education and Training Studies , 4 ( 10 ), 90 – 94 .

McMullen , M. G. ( 2009 ). Using language learning strategies to improve the writing skills of Saudi EFL students: Will it really work? System , 37 ( 3 ), 418 – 433 . doi: 10.1016/j.system.2009.05.001 .

Montero , C. R. , & Arizmendiarrieta , B. S. Y. ( 2017 ). The effectiveness of a learning strategies program for university students . Psicothema , 29 ( 4 ), 527 – 532 . doi: 10.7334/psicothema2016.171 .

Muelas , A. , & Navarro , E. ( 2015 ). Learning strategies and academic achievement . Procedia – Social and Behavioral Sciences , 165 , 217 – 221 . doi: 10.1016/j.sbspro.2014.12.625 , Proceeding in CPSYC 2014 .

Nikou , S. A. , & Economides , A. A. ( 2019 ). Mobile-based micro-learning and assessment: Impact on learning performance and motivation of high school students . Journal of Computer Assisted Learning , 34 ( 3 ), 269 – 278 . doi: 10.1111/jcal.12240 .

Pennequin , V. , Sorel , O. , Nanty , I. , & Fontaine , R. ( 2010 ). Metacognition and low achievement in mathematics: The effect of training in the use of metacognitive skills to solve mathematical word problems . Thinking and Reasoning , 16 ( 3 ), 198 – 220 . doi: 10.1080/13546783.2010.509052 .

Pinto , G. , Bigozzi , L. , Vettori , G. , & Vezzani , C. ( 2018 ). The relationship between conceptions of learning and academic outcomes in middle school students according to gender differences . Learning, Culture and Social Interactions , 16 , 45 – 54 . doi: 10.1016/j.lcsi.2017.11.001 .

Pordanjani , S. R. , & Guntur , L. M. ( 2019 ). Humanities investigating the implementation of critical literacy approach in the Middle-East education contexts: Three main constraints . ELS Journal on Interdisciplinary Studies on Humanities , 2 ( 3 ), 410 – 418 .

Quadlin , N. ( 2018 ). The mark of a woman's record: Gender and academic performance in hiring . American Sociological Review , 83 ( 2 ), 331 – 360 . doi: 10.1177/0003122418762291 .

Rosário , P. , Núñez , J. C. , Trigo , L. , Guimarães , C. , Fernández , E. , Cerezo , R. , Fuentes , S. , Orellana , M. , Santibáñez , A. , Fulano , C. , Ferreira , Â. , & Figueiredo , M. ( 2015 ). Transcultural analysis of the effectiveness of a program to promote self-regulated learning in Mozambique, Chile, Portugal, and Spain . Journal of Higher Education Research and Development , 34 ( 1 ), 173 – 187 . doi: 10.1080/07294360.2014.935932 .

Shehzad , M. W. , Razzaq , S. , Dahri , A. S. , & Shah , S. K. ( 2019 ). The association between reading self-efficacy beliefs and meta cognitive reading strategies among Saudi PYP students . The Dialogue , 14 ( 2 ), 32 – 43 .

Shi , H. ( 2017 ). Learning strategies and classification in education . Institute for Learning Styles Journal , 1 , 24 – 36 .

Tan , R. E. ( 2019 ). Academic self-concept, learning strategies and problem-solving achievement of university students . European Journal of Education Studies , 6 ( 2 ), 287 – 303 . doi: 10.5281/zenodo.3235652 .

Tariq , S. , Khan , M. , Afzal , S. , Shahzad , S. , Hamza , M. , Khan , H. , & Shaikh , S. ( 2016 ). Association between academic learning strategies and annual examination results among medical students of King Edward medical university . Annals of King Edward Medical University , 22 ( 2 ), 124 – 134 . doi: 10.21649/akemu.v22i2.1290 .

Tomar , S. , & Jindal , A. ( 2014 ). A study of effective learning strategies in relation to intelligence level across the science and arts academic streams of secondary level . IOSR Journal of Research and Method in Education (IOSR-JRME) , 4 ( 6 ), 41 – 50 .

Vega-Hernández , M. C. , Patino-Alonso , M. C. , Cabello , R. , Galindo-Villardón , M. P. , & Fernández-Berrocal , P. ( 2017 ). Perceived emotional intelligence and learning strategies in Spanish university students: A new perspective from a canonical non-symmetrical correspondence analysis . Front Psychology , 8 . doi: 10.3389/fpsyg.2017.01888 .

Vermunt , J. D. , & Vermunt , J. D. ( 2017 ). A learning patterns perspective on student learning in higher education: State of the art and moving forward . Educational Psychology Review , 29 ( 2 ), 269 – 299 .

Vettori , G. , Vezzani , C. , Bigozzi , L. , & Pinto , G. ( 2020 ). Upper secondary school students' conceptions of learning, learning strategies . The Journal of Educational Research , 113 ( 6 ), 475 – 485 . doi: 10.1080/00220671.2020.1861583 .

Further reading

Almusharraf , N. M. ( 2019 ). Learner autonomy and vocabulary development for Saudi university female EFL learners: Students' perspectives . International Journal of Linguistics , 11 ( 1 ), 166 – 195 .

Babbage , R. , Byers , R. , & Redding , H. ( 2008 ). Approaches to Teaching and Learning: Including Pupils Within Learning Difficulties . Oxen : David Fulton Publica .

Corresponding author

About the author.

Dr. Yousef Almoslamani is an Assistant Professor at the Instructional Technology Department, Faculty of Education, Ha'il University. He holds a PhD in Educational Technology from the University of Northern Colorado, USA.

Related articles

We’re listening — tell us what you think, something didn’t work….

Report bugs here

All feedback is valuable

Please share your general feedback

Join us on our journey

Platform update page.

Visit emeraldpublishing.com/platformupdate to discover the latest news and updates

Questions & More Information

Answers to the most commonly asked questions here

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 02 December 2020

Enhancing senior high school student engagement and academic performance using an inclusive and scalable inquiry-based program

• Locke Davenport Huyer   ORCID: orcid.org/0000-0003-1526-7122 1 , 2   na1 ,
• Neal I. Callaghan   ORCID: orcid.org/0000-0001-8214-3395 1 , 3   na1 ,
• Sara Dicks 4 ,
• Edward Scherer 4 ,
• Andrey I. Shukalyuk 1 ,
• Margaret Jou 4 &
• Dawn M. Kilkenny   ORCID: orcid.org/0000-0002-3899-9767 1 , 5  

npj Science of Learning volume  5 , Article number:  17 ( 2020 ) Cite this article

41k Accesses

4 Citations

13 Altmetric

Metrics details

The multi-disciplinary nature of science, technology, engineering, and math (STEM) careers often renders difficulty for high school students navigating from classroom knowledge to post-secondary pursuits. Discrepancies between the knowledge-based high school learning approach and the experiential approach of future studies leaves some students disillusioned by STEM. We present Discovery , a term-long inquiry-focused learning model delivered by STEM graduate students in collaboration with high school teachers, in the context of biomedical engineering. Entire classes of high school STEM students representing diverse cultural and socioeconomic backgrounds engaged in iterative, problem-based learning designed to emphasize critical thinking concomitantly within the secondary school and university environments. Assessment of grades and survey data suggested positive impact of this learning model on students’ STEM interests and engagement, notably in under-performing cohorts, as well as repeating cohorts that engage in the program on more than one occasion. Discovery presents a scalable platform that stimulates persistence in STEM learning, providing valuable learning opportunities and capturing cohorts of students that might otherwise be under-engaged in STEM.

Similar content being viewed by others

Subject integration and theme evolution of STEM education in K-12 and higher education research

Zehui Zhan & Shijing Niu

Skill levels and gains in university STEM education in China, India, Russia and the United States

Prashant Loyalka, Ou Lydia Liu, … Yanyan Li

Academic ecosystems must evolve to support a sustainable postdoc workforce

Murielle Ålund, Nathan Emery, … Eben Gering

Introduction

High school students with diverse STEM interests often struggle to understand the STEM experience outside the classroom 1 . The multi-disciplinary nature of many career fields can foster a challenge for students in their decision to enroll in appropriate high school courses while maintaining persistence in study, particularly when these courses are not mandatory 2 . Furthermore, this challenge is amplified by the known discrepancy between the knowledge-based learning approach common in high schools and the experiential, mastery-based approaches afforded by the subsequent undergraduate model 3 . In the latter, focused classes, interdisciplinary concepts, and laboratory experiences allow for the application of accumulated knowledge, practice in problem solving, and development of both general and technical skills 4 . Such immersive cooperative learning environments are difficult to establish in the secondary school setting and high school teachers often struggle to implement within their classroom 5 . As such, high school students may become disillusioned before graduation and never experience an enriched learning environment, despite their inherent interests in STEM 6 .

It cannot be argued that early introduction to varied math and science disciplines throughout high school is vital if students are to pursue STEM fields, especially within engineering 7 . However, the majority of literature focused on student interest and retention in STEM highlights outcomes in US high school learning environments, where the sciences are often subject-specific from the onset of enrollment 8 . In contrast, students in the Ontario (Canada) high school system are required to complete Level 1 and 2 core courses in science and math during Grades 9 and 10; these courses are offered as ‘applied’ or ‘academic’ versions and present broad topics of content 9 . It is not until Levels 3 and 4 (generally Grades 11 and 12, respectively) that STEM classes become subject-specific (i.e., Biology, Chemistry, and/or Physics) and are offered as “university”, “college”, or “mixed” versions, designed to best prepare students for their desired post-secondary pursuits 9 . Given that Levels 3 and 4 science courses are not mandatory for graduation, enrollment identifies an innate student interest in continued learning. Furthermore, engagement in these post-secondary preparatory courses is also dependent upon achieving successful grades in preceding courses, but as curriculum becomes more subject-specific, students often yield lower degrees of success in achieving course credit 2 . Therefore, it is imperative that learning supports are best focused on ensuring that those students with an innate interest are able to achieve success in learning.

When given opportunity and focused support, high school students are capable of successfully completing rigorous programs at STEM-focused schools 10 . Specialized STEM schools have existed in the US for over 100 years; generally, students are admitted after their sophomore year of high school experience (equivalent to Grade 10) based on standardized test scores, essays, portfolios, references, and/or interviews 11 . Common elements to this learning framework include a diverse array of advanced STEM courses, paired with opportunities to engage in and disseminate cutting-edge research 12 . Therein, said research experience is inherently based in the processes of critical thinking, problem solving, and collaboration. This learning framework supports translation of core curricular concepts to practice and is fundamental in allowing students to develop better understanding and appreciation of STEM career fields.

Despite the described positive attributes, many students do not have the ability or resources to engage within STEM-focused schools, particularly given that they are not prevalent across Canada, and other countries across the world. Consequently, many public institutions support the idea that post-secondary led engineering education programs are effective ways to expose high school students to engineering education and relevant career options, and also increase engineering awareness 13 . Although singular class field trips are used extensively to accomplish such programs, these may not allow immersive experiences for application of knowledge and practice of skills that are proven to impact long-term learning and influence career choices 14 , 15 . Longer-term immersive research experiences, such as after-school programs or summer camps, have shown successful at recruiting students into STEM degree programs and careers, where longevity of experience helps foster self-determination and interest-led, inquiry-based projects 4 , 16 , 17 , 18 , 19 .

Such activities convey the elements that are suggested to make a post-secondary led high school education programs successful: hands-on experience, self-motivated learning, real-life application, immediate feedback, and problem-based projects 20 , 21 . In combination with immersion in university teaching facilities, learning is authentic and relevant, similar to the STEM school-focused framework, and consequently representative of an experience found in actual STEM practice 22 . These outcomes may further be a consequence of student engagement and attitude: Brown et al. studied the relationships between STEM curriculum and student attitudes, and found the latter played a more important role in intention to persist in STEM when compared to self-efficacy 23 . This is interesting given that student self-efficacy has been identified to influence ‘motivation, persistence, and determination’ in overcoming challenges in a career pathway 24 . Taken together, this suggests that creation and delivery of modern, exciting curriculum that supports positive student attitudes is fundamental to engage and retain students in STEM programs.

Supported by the outcomes of identified effective learning strategies, University of Toronto (U of T) graduate trainees created a novel high school education program Discovery , to develop a comfortable yet stimulating environment of inquiry-focused iterative learning for senior high school students (Grades 11 & 12; Levels 3 & 4) at non-specialized schools. Built in strong collaboration with science teachers from George Harvey Collegiate Institute (Toronto District School Board), Discovery stimulates application of STEM concepts within a unique term-long applied curriculum delivered iteratively within both U of T undergraduate teaching facilities and collaborating high school classrooms 25 . Based on the volume of medically-themed news and entertainment that is communicated to the population at large, the rapidly-growing and diverse field of biomedical engineering (BME) were considered an ideal program context 26 . In its definition, BME necessitates cross-disciplinary STEM knowledge focused on the betterment of human health, wherein Discovery facilitates broadening student perspective through engaging inquiry-based projects. Importantly, Discovery allows all students within a class cohort to work together with their classroom teacher, stimulating continued development of a relevant learning community that is deemed essential for meaningful context and important for transforming student perspectives and understandings 27 , 28 . Multiple studies support the concept that relevant learning communities improve student attitudes towards learning, significantly increasing student motivation in STEM courses, and consequently improving the overall learning experience 29 . Learning communities, such as that provided by Discovery , also promote the formation of self-supporting groups, greater active involvement in class, and higher persistence rates for participating students 30 .

The objective of Discovery , through structure and dissemination, is to engage senior high school science students in challenging, inquiry-based practical BME activities as a mechanism to stimulate comprehension of STEM curriculum application to real-world concepts. Consequent focus is placed on critical thinking skill development through an atmosphere of perseverance in ambiguity, something not common in a secondary school knowledge-focused delivery but highly relevant in post-secondary STEM education strategies. Herein, we describe the observed impact of the differential project-based learning environment of Discovery on student performance and engagement. We identify the value of an inquiry-focused learning model that is tangible for students who struggle in a knowledge-focused delivery structure, where engagement in conceptual critical thinking in the relevant subject area stimulates student interest, attitudes, and resulting academic performance. Assessment of study outcomes suggests that when provided with a differential learning opportunity, student performance and interest in STEM increased. Consequently, Discovery provides an effective teaching and learning framework within a non-specialized school that motivates students, provides opportunity for critical thinking and problem-solving practice, and better prepares them for persistence in future STEM programs.

Program delivery

The outcomes of the current study result from execution of Discovery over five independent academic terms as a collaboration between Institute of Biomedical Engineering (graduate students, faculty, and support staff) and George Harvey Collegiate Institute (science teachers and administration) stakeholders. Each term, the program allowed senior secondary STEM students (Grades 11 and 12) opportunity to engage in a novel project-based learning environment. The program structure uses the problem-based engineering capstone framework as a tool of inquiry-focused learning objectives, motivated by a central BME global research topic, with research questions that are inter-related but specific to the curriculum of each STEM course subject (Fig. 1 ). Over each 12-week term, students worked in teams (3–4 students) within their class cohorts to execute projects with the guidance of U of T trainees ( Discovery instructors) and their own high school teacher(s). Student experimental work was conducted in U of T teaching facilities relevant to the research study of interest (i.e., Biology and Chemistry-based projects executed within Undergraduate Teaching Laboratories; Physics projects executed within Undergraduate Design Studios). Students were introduced to relevant techniques and safety procedures in advance of iterative experimentation. Importantly, this experience served as a course term project for students, who were assessed at several points throughout the program for performance in an inquiry-focused environment as well as within the regular classroom (Fig. 1 ). To instill the atmosphere of STEM, student teams delivered their outcomes in research poster format at a final symposium, sharing their results and recommendations with other post-secondary students, faculty, and community in an open environment.

The general program concept (blue background; top left ) highlights a global research topic examined through student dissemination of subject-specific research questions, yielding multifaceted student outcomes (orange background; top right ). Each program term (term workflow, yellow background; bottom panel ), students work on program deliverables in class (blue), iterate experimental outcomes within university facilities (orange), and are assessed accordingly at numerous deliverables in an inquiry-focused learning model.

Over the course of five terms there were 268 instances of tracked student participation, representing 170 individual students. Specifically, 94 students participated during only one term of programming, 57 students participated in two terms, 16 students participated in three terms, and 3 students participated in four terms. Multiple instances of participation represent students that enrol in more than one STEM class during their senior years of high school, or who participated in Grade 11 and subsequently Grade 12. Students were surveyed before and after each term to assess program effects on STEM interest and engagement. All grade-based assessments were performed by high school teachers for their respective STEM class cohorts using consistent grading rubrics and assignment structure. Here, we discuss the outcomes of student involvement in this experiential curriculum model.

Student performance and engagement

Student grades were assigned, collected, and anonymized by teachers for each Discovery deliverable (background essay, client meeting, proposal, progress report, poster, and final presentation). Teachers anonymized collective Discovery grades, the component deliverable grades thereof, final course grades, attendance in class and during programming, as well as incomplete classroom assignments, for comparative study purposes. Students performed significantly higher in their cumulative Discovery grade than in their cumulative classroom grade (final course grade less the Discovery contribution; p  < 0.0001). Nevertheless, there was a highly significant correlation ( p  < 0.0001) observed between the grade representing combined Discovery deliverables and the final course grade (Fig. 2a ). Further examination of the full dataset revealed two student cohorts of interest: the “Exceeds Expectations” (EE) subset (defined as those students who achieved ≥1 SD [18.0%] grade differential in Discovery over their final course grade; N  = 99 instances), and the “Multiple Term” (MT) subset (defined as those students who participated in Discovery more than once; 76 individual students that collectively accounted for 174 single terms of assessment out of the 268 total student-terms delivered) (Fig. 2b, c ). These subsets were not unrelated; 46 individual students who had multiple experiences (60.5% of total MTs) exhibited at least one occasion in achieving a ≥18.0% grade differential. As students participated in group work, there was concern that lower-performing students might negatively influence the Discovery grade of higher-performing students (or vice versa). However, students were observed to self-organize into groups where all individuals received similar final overall course grades (Fig. 2d ), thereby alleviating these concerns.

a Linear regression of student grades reveals a significant correlation ( p  = 0.0009) between Discovery performance and final course grade less the Discovery contribution to grade, as assessed by teachers. The dashed red line and intervals represent the theoretical 1:1 correlation between Discovery and course grades and standard deviation of the Discovery -course grade differential, respectively. b , c Identification of subgroups of interest, Exceeds Expectations (EE; N  = 99, orange ) who were ≥+1 SD in Discovery -course grade differential and Multi-Term (MT; N  = 174, teal ), of which N  = 65 students were present in both subgroups. d Students tended to self-assemble in working groups according to their final course performance; data presented as mean ± SEM. e For MT students participating at least 3 terms in Discovery , there was no significant correlation between course grade and time, while ( f ) there was a significant correlation between Discovery grade and cumulative terms in the program. Histograms of total absences per student in ( g ) Discovery and ( h ) class (binned by 4 days to be equivalent in time to a single Discovery absence).

The benefits experienced by MT students seemed progressive; MT students that participated in 3 or 4 terms ( N  = 16 and 3, respectively ) showed no significant increase by linear regression in their course grade over time ( p  = 0.15, Fig. 2e ), but did show a significant increase in their Discovery grades ( p  = 0.0011, Fig. 2f ). Finally, students demonstrated excellent Discovery attendance; at least 91% of participants attended all Discovery sessions in a given term (Fig. 2g ). In contrast, class attendance rates reveal a much wider distribution where 60.8% (163 out of 268 students) missed more than 4 classes (equivalent in learning time to one Discovery session) and 14.6% (39 out of 268 students) missed 16 or more classes (equivalent in learning time to an entire program of Discovery ) in a term (Fig. 2h ).

Discovery EE students (Fig. 3 ), roughly by definition, obtained lower course grades ( p  < 0.0001, Fig. 3a ) and higher final Discovery grades ( p  = 0.0004, Fig. 3b ) than non-EE students. This cohort of students exhibited program grades higher than classmates (Fig. 3c–h ); these differences were significant in every category with the exception of essays, where they outperformed to a significantly lesser degree ( p  = 0.097; Fig. 3c ). There was no statistically significant difference in EE vs. non-EE student classroom attendance ( p  = 0.85; Fig. 3i, j ). There were only four single day absences in Discovery within the EE subset; however, this difference was not statistically significant ( p  = 0.074).

The “Exceeds Expectations” (EE) subset of students (defined as those who received a combined Discovery grade ≥1 SD (18.0%) higher than their final course grade) performed ( a ) lower on their final course grade and ( b ) higher in the Discovery program as a whole when compared to their classmates. d – h EE students received significantly higher grades on each Discovery deliverable than their classmates, except for their ( c ) introductory essays and ( h ) final presentations. The EE subset also tended ( i ) to have a higher relative rate of attendance during Discovery sessions but no difference in ( j ) classroom attendance. N  = 99 EE students and 169 non-EE students (268 total). Grade data expressed as mean ± SEM.

Discovery MT students (Fig. 4 ), although not receiving significantly higher grades in class than students participating in the program only one time ( p  = 0.29, Fig. 4a ), were observed to obtain higher final Discovery grades than single-term students ( p  = 0.0067, Fig. 4b ). Although trends were less pronounced for individual MT student deliverables (Fig. 4c–h ), this student group performed significantly better on the progress report ( p  = 0.0021; Fig. 4f ). Trends of higher performance were observed for initial proposals and final presentations ( p  = 0.081 and 0.056, respectively; Fig. 4e, h ); all other deliverables were not significantly different between MT and non-MT students (Fig. 4c, d, g ). Attendance in Discovery ( p  = 0.22) was also not significantly different between MT and non-MT students, although MT students did miss significantly less class time ( p  = 0.010) (Fig. 4i, j ). Longitudinal assessment of individual deliverables for MT students that participated in three or more Discovery terms (Fig. 5 ) further highlights trend in improvement (Fig. 2f ). Greater performance over terms of participation was observed for essay ( p  = 0.0295, Fig. 5a ), client meeting ( p  = 0.0003, Fig. 5b ), proposal ( p  = 0.0004, Fig. 5c ), progress report ( p  = 0.16, Fig. 5d ), poster ( p  = 0.0005, Fig. 5e ), and presentation ( p  = 0.0295, Fig. 5f ) deliverable grades; these trends were all significant with the exception of the progress report ( p  = 0.16, Fig. 5d ) owing to strong performance in this deliverable in all terms.

The “multi-term” (MT) subset of students (defined as having attended more than one term of Discovery ) demonstrated favorable performance in Discovery , ( a ) showing no difference in course grade compared to single-term students, but ( b outperforming them in final Discovery grade. Independent of the number of times participating in Discovery , MT students did not score significantly differently on their ( c ) essay, ( d ) client meeting, or ( g ) poster. They tended to outperform their single-term classmates on the ( e ) proposal and ( h ) final presentation and scored significantly higher on their ( f ) progress report. MT students showed no statistical difference in ( i ) Discovery attendance but did show ( j ) higher rates of classroom attendance than single-term students. N  = 174 MT instances of student participation (76 individual students) and 94 single-term students. Grade data expressed as mean ± SEM.

Longitudinal assessment of a subset of MT student participants that participated in three ( N  = 16) or four ( N  = 3) terms presents a significant trend of improvement in their ( a ) essay, ( b ) client meeting, ( c ) proposal, ( e ) poster, and ( f ) presentation grade. d Progress report grades present a trend in improvement but demonstrate strong performance in all terms, limiting potential for student improvement. Grade data are presented as individual student performance; each student is represented by one color; data is fitted with a linear trendline (black).

Finally, the expansion of Discovery to a second school of lower LOI (i.e., nominally higher aggregate SES) allowed for the assessment of program impact in a new population over 2 terms of programming. A significant ( p  = 0.040) divergence in Discovery vs. course grade distribution from the theoretical 1:1 relationship was found in the new cohort (S 1 Appendix , Fig. S 1 ), in keeping with the pattern established in this study.

Teacher perceptions

Qualitative observation in the classroom by high school teachers emphasized the value students independently placed on program participation and deliverables. Throughout the term, students often prioritized Discovery group assignments over other tasks for their STEM courses, regardless of academic weight and/or due date. Comparing within this student population, teachers spoke of difficulties with late and incomplete assignments in the regular curriculum but found very few such instances with respect to Discovery -associated deliverables. Further, teachers speculated on the good behavior and focus of students in Discovery programming in contrast to attentiveness and behavior issues in their school classrooms. Multiple anecdotal examples were shared of renewed perception of student potential; students that exhibited poor academic performance in the classroom often engaged with high performance in this inquiry-focused atmosphere. Students appeared to take a sense of ownership, excitement, and pride in the setting of group projects oriented around scientific inquiry, discovery, and dissemination.

Student perceptions

Students were asked to consider and rank the academic difficulty (scale of 1–5, with 1 = not challenging and 5 = highly challenging) of the work they conducted within the Discovery learning model. Considering individual Discovery terms, at least 91% of students felt the curriculum to be sufficiently challenging with a 3/5 or higher ranking (Term 1: 87.5%, Term 2: 93.4%, Term 3: 85%, Term 4: 93.3%, Term 5: 100%), and a minimum of 58% of students indicating a 4/5 or higher ranking (Term 1: 58.3%, Term 2: 70.5%, Term 3: 67.5%, Term 4: 69.1%, Term 5: 86.4%) (Fig. 6a ).

a Histogram of relative frequency of perceived Discovery programming academic difficulty ranked from not challenging (1) to highly challenging (5) for each session demonstrated the consistently perceived high degree of difficulty for Discovery programming (total responses: 223). b Program participation increased student comfort (94.6%) with navigating lab work in a university or college setting (total responses: 220). c Considering participation in Discovery programming, students indicated their increased (72.4%) or decreased (10.1%) likelihood to pursue future experiences in STEM as a measure of program impact (total responses: 217). d Large majority of participating students (84.9%) indicated their interest for future participation in Discovery (total responses: 212). Students were given the opportunity to opt out of individual survey questions, partially completed surveys were included in totals.

The majority of students (94.6%) indicated they felt more comfortable with the idea of performing future work in a university STEM laboratory environment given exposure to university teaching facilities throughout the program (Fig. 6b ). Students were also queried whether they were (i) more likely, (ii) less likely, or (iii) not impacted by their experience in the pursuit of STEM in the future. The majority of participants (>82%) perceived impact on STEM interests, with 72.4% indicating they were more likely to pursue these interests in the future (Fig. 6c ). When surveyed at the end of term, 84.9% of students indicated they would participate in the program again (Fig. 6d ).

We have described an inquiry-based framework for implementing experiential STEM education in a BME setting. Using this model, we engaged 268 instances of student participation (170 individual students who participated 1–4 times) over five terms in project-based learning wherein students worked in peer-based teams under the mentorship of U of T trainees to design and execute the scientific method in answering a relevant research question. Collaboration between high school teachers and Discovery instructors allowed for high school student exposure to cutting-edge BME research topics, participation in facilitated inquiry, and acquisition of knowledge through scientific discovery. All assessments were conducted by high school teachers and constituted a fraction (10–15%) of the overall course grade, instilling academic value for participating students. As such, students exhibited excitement to learn as well as commitment to their studies in the program.

Through our observations and analysis, we suggest there is value in differential learning environments for students that struggle in a knowledge acquisition-focused classroom setting. In general, we observed a high level of academic performance in Discovery programming (Fig. 2a ), which was highlighted exceptionally in EE students who exhibited greater academic performance in Discovery deliverables compared to normal coursework (>18% grade improvement in relevant deliverables). We initially considered whether this was the result of strong students influencing weaker students; however, group organization within each course suggests this is not the case (Fig. 2d ). With the exception of one class in one term (24 participants assigned by their teacher), students were allowed to self-organize into working groups and they chose to work with other students of relatively similar academic performance (as indicated by course grade), a trend observed in other studies 31 , 32 . Remarkably, EE students not only excelled during Discovery when compared to their own performance in class, but this cohort also achieved significantly higher average grades in each of the deliverables throughout the program when compared to the remaining Discovery cohort (Fig. 3 ). This data demonstrates the value of an inquiry-based learning environment compared to knowledge-focused delivery in the classroom in allowing students to excel. We expect that part of this engagement was resultant of student excitement with a novel learning opportunity. It is however a well-supported concept that students who struggle in traditional settings tend to demonstrate improved interest and motivation in STEM when given opportunity to interact in a hands-on fashion, which supports our outcomes 4 , 33 . Furthermore, these outcomes clearly represent variable student learning styles, where some students benefit from a greater exchange of information, knowledge and skills in a cooperative learning environment 34 . The performance of the EE group may not be by itself surprising, as the identification of the subset by definition required high performers in Discovery who did not have exceptionally high course grades; in addition, the final Discovery grade is dependent on the component assignment grades. However, the discrepancies between EE and non-EE groups attendance suggests that students were engaged by Discovery in a way that they were not by regular classroom curriculum.

In addition to quantified engagement in Discovery observed in academic performance, we believe remarkable attendance rates are indicative of the value students place in the differential learning structure. Given the differences in number of Discovery days and implications of missing one day of regular class compared to this immersive program, we acknowledge it is challenging to directly compare attendance data and therefore approximate this comparison with consideration of learning time equivalence. When combined with other subjective data including student focus, requests to work on Discovery during class time, and lack of discipline/behavior issues, the attendance data importantly suggests that students were especially engaged by the Discovery model. Further, we believe the increased commute time to the university campus (students are responsible for independent transit to campus, a much longer endeavour than the normal school commute), early program start time, and students’ lack of familiarity with the location are non-trivial considerations when determining the propensity of students to participate enthusiastically in Discovery . We feel this suggests the students place value on this team-focused learning and find it to be more applicable and meaningful to their interests.

Given post-secondary admission requirements for STEM programs, it would be prudent to think that students participating in multiple STEM classes across terms are the ones with the most inherent interest in post-secondary STEM programs. The MT subset, representing students who participated in Discovery for more than one term, averaged significantly higher final Discovery grades. The increase in the final Discovery grade was observed to result from a general confluence of improved performance over multiple deliverables and a continuous effort to improve in a STEM curriculum. This was reflected in longitudinal tracking of Discovery performance, where we observed a significant trend of improved performance. Interestingly, the high number of MT students who were included in the EE group suggests that students who had a keen interest in science enrolled in more than one course and in general responded well to the inquiry-based teaching method of Discovery , where scientific method was put into action. It stands to reason that students interested in science will continue to take STEM courses and will respond favorably to opportunities to put classroom theory to practical application.

The true value of an inquiry-based program such as Discovery may not be based in inspiring students to perform at a higher standard in STEM within the high school setting, as skills in critical thinking do not necessarily translate to knowledge-based assessment. Notably, students found the programming equally challenging throughout each of the sequential sessions, perhaps somewhat surprising considering the increasing number of repeat attendees in successive sessions (Fig. 6a ). Regardless of sub-discipline, there was an emphasis of perceived value demonstrated through student surveys where we observed indicated interest in STEM and comfort with laboratory work environments, and desire to engage in future iterations given the opportunity. Although non-quantitative, we perceive this as an indicator of significant student engagement, even though some participants did not yield academic success in the program and found it highly challenging given its ambiguity.

Although we observed that students become more certain of their direction in STEM, further longitudinal study is warranted to make claim of this outcome. Additionally, at this point in our assessment we cannot effectively assess the practical outcomes of participation, understanding that the immediate effects observed are subject to a number of factors associated with performance in the high school learning environment. Future studies that track graduates from this program will be prudent, in conjunction with an ever-growing dataset of assessment as well as surveys designed to better elucidate underlying perceptions and attitudes, to continue to understand the expected benefits of this inquiry-focused and partnered approach. Altogether, a multifaceted assessment of our early outcomes suggests significant value of an immersive and iterative interaction with STEM as part of the high school experience. A well-defined divergence from knowledge-based learning, focused on engagement in critical thinking development framed in the cutting-edge of STEM, may be an important step to broadening student perspectives.

In this study, we describe the short-term effects of an inquiry-based STEM educational experience on a cohort of secondary students attending a non-specialized school, and suggest that the framework can be widely applied across virtually all subjects where inquiry-driven and mentored projects can be undertaken. Although we have demonstrated replication in a second cohort of nominally higher SES (S 1 Appendix , Supplementary Fig. 1 ), a larger collection period with more students will be necessary to conclusively determine impact independent of both SES and specific cohort effects. Teachers may also find this framework difficult to implement depending on resources and/or institutional investment and support, particularly if post-secondary collaboration is inaccessible. Offerings to a specific subject (e.g., physics) where experiments yielding empirical data are logistically or financially simpler to perform may be valid routes of adoption as opposed to the current study where all subject cohorts were included.

As we consider Discovery in a bigger picture context, expansion and implementation of this model is translatable. Execution of the scientific method is an important aspect of citizen science, as the concepts of critical thing become ever-more important in a landscape of changing technological landscapes. Giving students critical thinking and problem-solving skills in their primary and secondary education provides value in the context of any career path. Further, we feel that this model is scalable across disciplines, STEM or otherwise, as a means of building the tools of inquiry. We have observed here the value of differential inclusive student engagement and critical thinking through an inquiry-focused model for a subset of students, but further to this an engagement, interest, and excitement across the body of student participants. As we educate the leaders of tomorrow, we suggest that use of an inquiry-focused model such as Discovery could facilitate growth of a data-driven critical thinking framework.

In conclusion, we have presented a model of inquiry-based STEM education for secondary students that emphasizes inclusion, quantitative analysis, and critical thinking. Student grades suggest significant performance benefits, and engagement data suggests positive student attitude despite the perceived challenges of the program. We also note a particular performance benefit to students who repeatedly engage in the program. This framework may carry benefits in a wide variety of settings and disciplines for enhancing student engagement and performance, particularly in non-specialized school environments.

Study design and implementation

Participants in Discovery include all students enrolled in university-stream Grade 11 or 12 biology, chemistry, or physics at the participating school over five consecutive terms (cohort summary shown in Table 1 ). Although student participation in educational content was mandatory, student grades and survey responses (administered by high school teachers) were collected from only those students with parent or guardian consent. Teachers replaced each student name with a unique coded identifier to preserve anonymity but enable individual student tracking over multiple terms. All data collected were analyzed without any exclusions save for missing survey responses; no power analysis was performed prior to data collection.

Ethics statement

This study was approved by the University of Toronto Health Sciences Research Ethics Board (Protocol # 34825) and the Toronto District School Board External Research Review Committee (Protocol # 2017-2018-20). Written informed consent was collected from parents or guardians of participating students prior to the acquisition of student data (both post-hoc academic data and survey administration). Data were anonymized by high school teachers for maintenance of academic confidentiality of individual students prior to release to U of T researchers.

Educational program overview

Students enrolled in university-preparatory STEM classes at the participating school completed a term-long project under the guidance of graduate student instructors and undergraduate student mentors as a mandatory component of their respective course. Project curriculum developed collaboratively between graduate students and participating high school teachers was delivered within U of T Faculty of Applied Science & Engineering (FASE) teaching facilities. Participation allows high school students to garner a better understanding as to how undergraduate learning and career workflows in STEM vary from traditional high school classroom learning, meanwhile reinforcing the benefits of problem solving, perseverance, teamwork, and creative thinking competencies. Given that Discovery was a mandatory component of course curriculum, students participated as class cohorts and addressed questions specific to their course subject knowledge base but related to the defined global health research topic (Fig. 1 ). Assessment of program deliverables was collectively assigned to represent 10–15% of the final course grade for each subject at the discretion of the respective STEM teacher.

The Discovery program framework was developed, prior to initiation of student assessment, in collaboration with one high school selected from the local public school board over a 1.5 year period of time. This partner school consistently scores highly (top decile) in the school board’s Learning Opportunities Index (LOI). The LOI ranks each school based on measures of external challenges affecting its student population therefore schools with the greatest level of external challenge receive a higher ranking 35 . A high LOI ranking is inversely correlated with socioeconomic status (SES); therefore, participating students are identified as having a significant number of external challenges that may affect their academic success. The mandatory nature of program participation was established to reach highly capable students who may be reluctant to engage on their own initiative, as a means of enhancing the inclusivity and impact of the program. The selected school partner is located within a reasonable geographical radius of our campus (i.e., ~40 min transit time from school to campus). This is relevant as participating students are required to independently commute to campus for Discovery hands-on experiences.

Each program term of Discovery corresponds with a five-month high school term. Lead university trainee instructors (3–6 each term) engaged with high school teachers 1–2 months in advance of high school student engagement to discern a relevant overarching global healthcare theme. Each theme was selected with consideration of (a) topics that university faculty identify as cutting-edge biomedical research, (b) expertise that Discovery instructors provide, and (c) capacity to showcase the diversity of BME. Each theme was sub-divided into STEM subject-specific research questions aligning with provincial Ministry of Education curriculum concepts for university-preparatory Biology, Chemistry, and Physics 9 that students worked to address, both on-campus and in-class, during a term-long project. The Discovery framework therefore provides students a problem-based learning experience reflective of an engineering capstone design project, including a motivating scientific problem (i.e., global topic), subject-specific research question, and systematic determination of a professional recommendation addressing the needs of the presented problem.

Discovery instructors were volunteers recruited primarily from graduate and undergraduate BME programs in the FASE. Instructors were organized into subject-specific instructional teams based on laboratory skills, teaching experience, and research expertise. The lead instructors of each subject (the identified 1–2 trainees that built curriculum with high school teachers) were responsible to organize the remaining team members as mentors for specific student groups over the course of the program term (~1:8 mentor to student ratio).

All Discovery instructors were familiarized with program expectations and trained in relevant workspace safety, in addition to engagement at a teaching workshop delivered by the Faculty Advisor (a Teaching Stream faculty member) at the onset of term. This workshop was designed to provide practical information on teaching and was co-developed with high school teachers based on their extensive training and experience in fundamental teaching methods. In addition, group mentors received hands-on training and guidance from lead instructors regarding the specific activities outlined for their respective subject programming (an exemplary term of student programming is available in S 2 Appendix) .

Discovery instructors were responsible for introducing relevant STEM skills and mentoring high school students for the duration of their projects, with support and mentorship from the Faculty Mentor. Each instructor worked exclusively throughout the term with the student groups to which they had been assigned, ensuring consistent mentorship across all disciplinary components of the project. In addition to further supporting university trainees in on-campus mentorship, high school teachers were responsible for academic assessment of all student program deliverables (Fig. 1 ; the standardized grade distribution available in S 3 Appendix ). Importantly, trainees never engaged in deliverable assessment; for continuity of overall course assessment, this remained the responsibility of the relevant teacher for each student cohort.

Throughout each term, students engaged within the university facilities four times. The first three sessions included hands-on lab sessions while the fourth visit included a culminating symposium for students to present their scientific findings (Fig. 1 ). On average, there were 4–5 groups of students per subject (3–4 students per group; ~20 students/class). Discovery instructors worked exclusively with 1–2 groups each term in the capacity of mentor to monitor and guide student progress in all project deliverables.

After introducing the selected global research topic in class, teachers led students in completion of background research essays. Students subsequently engaged in a subject-relevant skill-building protocol during their first visit to university teaching laboratory facilities, allowing opportunity to understand analysis techniques and equipment relevant for their assessment projects. At completion of this session, student groups were presented with a subject-specific research question as well as the relevant laboratory inventory available for use during their projects. Armed with this information, student groups continued to work in their classroom setting to develop group-specific experimental plans. Teachers and Discovery instructors provided written and oral feedback, respectively , allowing students an opportunity to revise their plans in class prior to on-campus experimental execution.

Once at the relevant laboratory environment, student groups executed their protocols in an effort to collect experimental data. Data analysis was performed in the classroom and students learned by trial & error to optimize their protocols before returning to the university lab for a second opportunity of data collection. All methods and data were re-analyzed in class in order for students to create a scientific poster for the purpose of study/experience dissemination. During a final visit to campus, all groups presented their findings at a research symposium, allowing students to verbally defend their process, analyses, interpretations, and design recommendations to a diverse audience including peers, STEM teachers, undergraduate and graduate university students, postdoctoral fellows and U of T faculty.

Data collection

Teachers evaluated their students on the following associated deliverables: (i) global theme background research essay; (ii) experimental plan; (iii) progress report; (iv) final poster content and presentation; and (v) attendance. For research purposes, these grades were examined individually and also as a collective Discovery program grade for each student. For students consenting to participation in the research study, all Discovery grades were anonymized by the classroom teacher before being shared with study authors. Each student was assigned a code by the teacher for direct comparison of deliverable outcomes and survey responses. All instances of “Final course grade” represent the prorated course grade without the Discovery component, to prevent confounding of quantitative analyses.

Survey instruments were used to gain insight into student attitudes and perceptions of STEM and post-secondary study, as well as Discovery program experience and impact (S 4 Appendix ). High school teachers administered surveys in the classroom only to students supported by parental permission. Pre-program surveys were completed at minimum 1 week prior to program initiation each term and exit surveys were completed at maximum 2 weeks post- Discovery term completion. Surveys results were validated using a principal component analysis (S 1 Appendix , Supplementary Fig. 2 ).

Identification and comparison of population subsets

From initial analysis, we identified two student subpopulations of particular interest: students who performed ≥1 SD [18.0%] or greater in the collective Discovery components of the course compared to their final course grade (“EE”), and students who participated in Discovery more than once (“MT”). These groups were compared individually against the rest of the respective Discovery population (“non-EE” and “non-MT”, respectively ). Additionally, MT students who participated in three or four (the maximum observed) terms of Discovery were assessed for longitudinal changes to performance in their course and Discovery grades. Comparisons were made for all Discovery deliverables (introductory essay, client meeting, proposal, progress report, poster, and presentation), final Discovery grade, final course grade, Discovery attendance, and overall attendance.

Statistical analysis

Student course grades were analyzed in all instances without the Discovery contribution (calculated from all deliverable component grades and ranging from 10 to 15% of final course grade depending on class and year) to prevent correlation. Aggregate course grades and Discovery grades were first compared by paired t-test, matching each student’s course grade to their Discovery grade for the term. Student performance in Discovery ( N  = 268 instances of student participation, comprising 170 individual students that participated 1–4 times) was initially assessed in a linear regression of Discovery grade vs. final course grade. Trends in course and Discovery performance over time for students participating 3 or 4 terms ( N  = 16 and 3 individuals, respectively ) were also assessed by linear regression. For subpopulation analysis (EE and MT, N  = 99 instances from 81 individuals and 174 instances from 76 individuals, respectively ), each dataset was tested for normality using the D’Agostino and Pearson omnibus normality test. All subgroup comparisons vs. the remaining population were performed by Mann–Whitney U -test. Data are plotted as individual points with mean ± SEM overlaid (grades), or in histogram bins of 1 and 4 days, respectively , for Discovery and class attendance. Significance was set at α ≤ 0.05.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

The data that support the findings of this study are available upon reasonable request from the corresponding author DMK. These data are not publicly available due to privacy concerns of personal data according to the ethical research agreements supporting this study.

Holmes, K., Gore, J., Smith, M. & Lloyd, A. An integrated analysis of school students’ aspirations for STEM careers: Which student and school factors are most predictive? Int. J. Sci. Math. Educ. 16 , 655–675 (2018).

Article   Google Scholar  

Dooley, M., Payne, A., Steffler, M. & Wagner, J. Understanding the STEM path through high school and into university programs. Can. Public Policy 43 , 1–16 (2017).

Gilmore, M. W. Improvement of STEM education: experiential learning is the key. Mod. Chem. Appl. 1, e109. https://doi.org/10.4172/2329-6798.1000e109 (2013).

Roberts, T. et al. Students’ perceptions of STEM learning after participating in a summer informal learning experience. Int. J. STEM Educ. 5 , 35 (2018).

Gillies, R. M. & Boyle, M. Teachers’ reflections on cooperative learning: Issues of implementation. Teach. Teach. Educ. 26 , 933–940 (2010).

Nasir, M., Seta, J. & Meyer, E.G. Introducing high school students to biomedical engineering through summer camps. Paper presented at the ASEE Annual Conference & Exposition, Indianapolis, IN. https://doi.org/10.18260/1-2-20701 (2014).

Sadler, P. M., Sonnert, G., Hazari, Z. & Tai, R. Stability and volatility of STEM career interest in high school: a gender study. Sci. Educ. 96 , 411–427 (2012).

Sarikas, C. The High School Science Classes You Should Take . https://blog.prepscholar.com/the-high-school-science-classes-you-should-take (2020).

Ontario, G. o. The ontario curriculum grades 11 and 12. Science http://www.edu.gov.on.ca/eng/curriculum/secondary/2009science11_12.pdf (2008).

Scott, C. An investigation of science, technology, engineering and mathematics (STEM) focused high schools in the US. J. STEM Educ.: Innov. Res. 13 , 30 (2012).

Google Scholar  

Erdogan, N. & Stuessy, C. L. Modeling successful STEM high schools in the United States: an ecology framework. Int. J. Educ. Math., Sci. Technol. 3 , 77–92 (2015).

Pfeiffer, S. I., Overstreet, J. M. & Park, A. The state of science and mathematics education in state-supported residential academies: a nationwide survey. Roeper Rev. 32 , 25–31 (2009).

Anthony, A. B., Greene, H., Post, P. E., Parkhurst, A. & Zhan, X. Preparing university students to lead K-12 engineering outreach programmes: a design experiment. Eur. J. Eng. Educ. 41 , 623–637 (2016).

Brown, J. S., Collins, A. & Duguid, P. Situated cognition and the culture of learning. Educ. researcher 18 , 32–42 (1989).

Reveles, J. M. & Brown, B. A. Contextual shifting: teachers emphasizing students’ academic identity to promote scientific literacy. Sci. Educ. 92 , 1015–1041 (2008).

Adedokun, O. A., Bessenbacher, A. B., Parker, L. C., Kirkham, L. L. & Burgess, W. D. Research skills and STEM undergraduate research students’ aspirations for research careers: mediating effects of research self-efficacy. J. Res. Sci. Teach. 50 , 940–951 (2013).

Boekaerts, M. Self-regulated learning: a new concept embraced by researchers, policy makers, educators, teachers, and students. Learn. Instr. 7 , 161–186 (1997).

Honey, M., Pearson, G. & Schweingruber, H. STEM Integration in K-12 Education: Status, Prospects, and An Agenda for Research . (National Academies Press, Washington, DC, 2014).

Moote, J. K., Williams, J. M. & Sproule, J. When students take control: investigating the impact of the crest inquiry-based learning program on self-regulated processes and related motivations in young science students. J. Cogn. Educ. Psychol. 12 , 178–196 (2013).

Fantz, T. D., Siller, T. J. & Demiranda, M. A. Pre-collegiate factors influencing the self-efficacy of engineering students. J. Eng. Educ. 100 , 604–623 (2011).

Ralston, P. A., Hieb, J. L. & Rivoli, G. Partnerships and experience in building STEM pipelines. J. Professional Issues Eng. Educ. Pract. 139 , 156–162 (2012).

Kelley, T. R. & Knowles, J. G. A conceptual framework for integrated STEM education. Int. J. STEM Educ. 3 , 11 (2016).

Brown, P. L., Concannon, J. P., Marx, D., Donaldson, C. W. & Black, A. An examination of middle school students’ STEM self-efficacy with relation to interest and perceptions of STEM. J. STEM Educ.: Innov. Res. 17 , 27–38 (2016).

Bandura, A., Barbaranelli, C., Caprara, G. V. & Pastorelli, C. Self-efficacy beliefs as shapers of children’s aspirations and career trajectories. Child Dev. 72 , 187–206 (2001).

Article   CAS   Google Scholar  

Davenport Huyer, L. et al. IBBME discovery: biomedical engineering-based iterative learning in a high school STEM curriculum (evaluation). Paper presented at ASEE Annual Conference & Exposition, Salt Lake City, UT. https://doi.org/10.18260/1-2-30591 (2018).

Abu-Faraj, Ziad O., ed. Handbook of research on biomedical engineering education and advanced bioengineering learning: interdisciplinary concepts: interdisciplinary concepts. Vol. 2. IGI Global (2012).

Johri, A. & Olds, B. M. Situated engineering learning: bridging engineering education research and the learning sciences. J. Eng. Educ. 100 , 151–185 (2011).

O’Connell, K. B., Keys, B. & Storksdieck, M. Taking stock of oregon STEM hubs: accomplishments and challenges. Corvallis: Oregon State University https://ir.library.oregonstate.edu/concern/articles/hq37vt23t (2017).

Freeman, K. E., Alston, S. T. & Winborne, D. G. Do learning communities enhance the quality of students’ learning and motivation in STEM? J. Negro Educ. 77 , 227–240 (2008).

Weaver, R. R. & Qi, J. Classroom organization and participation: college students’ perceptions. J. High. Educ. 76 , 570–601 (2005).

Chapman, K. J., Meuter, M., Toy, D. & Wright, L. Can’t we pick our own groups? The influence of group selection method on group dynamics and outcomes. J. Manag. Educ. 30 , 557–569 (2006).

Hassaskhah, J. & Mozaffari, H. The impact of group formation method (student-selected vs. teacher-assigned) on group dynamics and group outcome in EFL creative writing. J. Lang. Teach. Res. 6 , 147–156 (2015).

Ma, V. J. & Ma, X. A comparative analysis of the relationship between learning styles and mathematics performance. Int. J. STEM Educ. 1 , 3 (2014).

Weinstein, C. E. & Hume, L. M. Study Strategies for Lifelong Learning . (American Psychological Association, 1998).

Toronto District School Board. The 2017 Learning Opportunities Index: Questions and Answers. https://www.tdsb.on.ca/Portals/research/docs/reports/LOI2017v2.pdf (2017).

Download references

Acknowledgements

This study has been possible due to the support of many University of Toronto trainee volunteers, including Genevieve Conant, Sherif Ramadan, Daniel Smieja, Rami Saab, Andrew Effat, Serena Mandla, Cindy Bui, Janice Wong, Dawn Bannerman, Allison Clement, Shouka Parvin Nejad, Nicolas Ivanov, Jose Cardenas, Huntley Chang, Romario Regeenes, Dr. Henrik Persson, Ali Mojdeh, Nhien Tran-Nguyen, Ileana Co, and Jonathan Rubianto. We further acknowledge the staff and administration of George Harvey Collegiate Institute and the Institute of Biomedical Engineering (IBME), as well as Benjamin Rocheleau and Madeleine Rocheleau for contributions to data collation. Discovery has grown with continued support of Dean Christopher Yip (Faculty of Applied Science and Engineering, U of T), and the financial support of the IBME and the National Science and Engineering Research Council (NSERC) PromoScience program (PROSC 515876-2017; IBME “Igniting Youth Curiosity in STEM” initiative co-directed by DMK and Dr. Penney Gilbert). LDH and NIC were supported by Vanier Canada graduate scholarships from the Canadian Institutes of Health Research and NSERC, respectively . DMK holds a Dean’s Emerging Innovation in Teaching Professorship in the Faculty of Engineering & Applied Science, U of T.

Author information

These authors contributed equally: Locke Davenport Huyer, Neal I. Callaghan.

Authors and Affiliations

Institute of Biomedical Engineering, University of Toronto, Toronto, ON, Canada

Locke Davenport Huyer, Neal I. Callaghan, Andrey I. Shukalyuk & Dawn M. Kilkenny

Chemical Engineering and Applied Chemistry, University of Toronto, Toronto, ON, Canada

Locke Davenport Huyer

Translational Biology and Engineering Program, Ted Rogers Centre for Heart Research, University of Toronto, Toronto, ON, Canada

Neal I. Callaghan

George Harvey Collegiate Institute, Toronto District School Board, Toronto, ON, Canada

Sara Dicks, Edward Scherer & Margaret Jou

Institute for Studies in Transdisciplinary Engineering Education & Practice, University of Toronto, Toronto, ON, Canada

Dawn M. Kilkenny

You can also search for this author in PubMed   Google Scholar

Contributions

LDH, NIC and DMK conceived the program structure, designed the study, and interpreted the data. LDH and NIC ideated programming, coordinated execution, and performed all data analysis. SD, ES, and MJ designed and assessed student deliverables, collected data, and anonymized data for assessment. SD assisted in data interpretation. AIS assisted in programming ideation and design. All authors provided feedback and approved the manuscript that was written by LDH, NIC and DMK.

Corresponding author

Correspondence to Dawn M. Kilkenny .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplemental material, reporting summary, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Davenport Huyer, L., Callaghan, N.I., Dicks, S. et al. Enhancing senior high school student engagement and academic performance using an inclusive and scalable inquiry-based program. npj Sci. Learn. 5 , 17 (2020). https://doi.org/10.1038/s41539-020-00076-2

Download citation

Received : 05 December 2019

Accepted : 08 October 2020

Published : 02 December 2020

DOI : https://doi.org/10.1038/s41539-020-00076-2

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Front Psychol

How and Why Do Students Use Learning Strategies? A Mixed Methods Study on Learning Strategies and Desirable Difficulties With Effective Strategy Users

Associated data.

In order to ensure long-term retention of information students must move from relying on surface-level approaches that are seemingly effective in the short-term to “building in” so called “desirable difficulties,” with the aim of achieving understanding and long-term retention of the subject matter. But how can this level of self-regulation be achieved by students when learning? Traditionally, research on learning strategy use is performed using self-report questionnaires. As this method is accompanied by several drawbacks, we chose a qualitative, in-depth approach to inquire about students' strategies and to investigate how students successfully self-regulate their learning. In order to paint a picture of effective learning strategy use, focus groups were organized in which previously identified, effectively self-regulating students ( N = 26) were asked to explain how they approach their learning. Using a constructivist grounded theory methodology, a model was constructed describing how effective strategy users manage their learning. In this model, students are driven by a personal learning goal, adopting a predominantly qualitative, or quantitative approach to learning. While learning, students are continually engaged in active processing and self-monitoring. This process is guided by a constant balancing between adhering to established study habits, while maintaining a sufficient degree of flexibility to adapt to changes in the learning environment, assessment demands, and time limitations. Indeed, students reported using several strategies, some of which are traditionally regarded as “ineffective” (highlighting, rereading etc.). However, they used them in a way that fit their learning situation. Implications are discussed for the incorporation of desirable difficulties in higher education.

Introduction

Self-regulated learning (SRL) refers to the “process whereby students activate and sustain cognitions, behaviors and affects, which are systematically oriented toward the attainment of their goals” [(Schunk and Zimmerman, 1994 ), p. 309]. With the enormous increase in available information since the emergence of the Internet (Arbesman, 2013 ), SRL is becoming increasingly important in modern education. This can be especially daunting for students in a problem-based curriculum, as this approach places high demands on students' independent self-study and individual search for information (e.g., Kirschner et al., 2006 ). Students will need effective self-regulatory strategies in order to successfully navigate this educational landscape. As students often rely on ineffective, surface-level study strategies (Kornell and Bjork, 2007 ), it is important to understand what constitutes effective strategy use in a problem-based curriculum, and how to improve SRL in students not skilled in self-regulation.

An important concept in this regard is that of “desirable difficulties.” What constitutes as “desirable” when introducing difficulties into the learning process, at least from the students' perspective, will likely depend on the goals they set for learning. Learning goals can include long-term understanding and transfer, or simply a desire to pass an exam. When the aim is simply to pass the test, different learning strategies apply than when the focus is on long-term understanding and transfer. In fact, strategies which have a positive effect on long-term understanding and transfer, may even have a negative effect on learning in the short term and vice versa (Van Merriënboer et al., 1997 ; Helsdingen et al., 2011 ; Van Merriënboer and Kirschner, 2018 ). However, this short-term achievement will not prepare students for long-term, professional practice (Boud and Falchikov, 2006 ). From an educational perspective, the focus should therefore be on long-term retention and transfer. Indeed, as defined by Bjork ( 1994 ), creating desirable difficulties when learning refers to the process in which students use effortful learning strategies, with the aim of achieving long-term learning benefits, rather than surface-level strategies which are only effective in the short-term.

The traditional way of measuring students' strategy use is through self-report surveys (Panadero et al., 2016 ). These studies often reveal that students rely on ineffective strategies when studying. For example, Blasiman et al. ( 2017 ) found that over the course of a semester, students often relied on ineffective strategies such as reading notes and rereading text. Similarly, Karpicke et al. ( 2009 ) found that while students often rely on rereading strategies, few students use more effective strategies like retrieval practice. One of the drawbacks of this form of measurement is that students are usually confronted with a set of predefined strategy categories to choose from. Authors have raised questions about whether self-report questionnaires are able to gauge students' use of different learning strategies across different contexts and tasks (Winne and Hadwin, 1998 ; Perry and Winne, 2006 ; Schellings, 2011 ; McCardle and Hadwin, 2015 ), students' ability to recover the required information from their memory (Perry and Winne, 2006 ), the possibility of socially desirable answers (Bråten and Samuelstuen, 2007 ), and a potential tendency for students to rate the value they attach to a certain strategy rather than their actual use (Bråten and Samuelstuen, 2007 ; Bernacki et al., 2012 ). Another possibility is that students use certain strategies to regulate their learning which they do not recognize as belonging to a particular category (Veenman, 2011 ). Furthermore, it is possible that strategies which are traditionally treated as ineffective by these self-report questionnaires are in fact adapted by students to fit their learning situation and goals in an effective way. These expectations were the basis for exploration in the current study.

In order to overcome these difficulties, a more qualitative, in-depth approach to inquiring about students' use of learning strategies can be worthwhile in order to investigate how students successfully self-regulate their learning. Specifically, this rich form of data collection allows for the description of different contexts and learning tasks, allowing students to distinguish between different learning strategies used in different situations and for different goals, as well as how they potentially use seemingly “ineffective” strategies to adapt to a learning situation or goal. A qualitative approach to inquiry enables students to give more elaborate explanations for as to how and why they use particular strategies, as well as potential variations with regard to varying circumstances. By carefully constructing the questions, it should also be possible to distinguish between the value students attach to different strategies vs. their actual use. Furthermore, students' rich descriptions of their approaches to learning allow the researcher to identify strategies that students would be unable to correctly label in a questionnaire.

As a qualitative approach to data collection, the focus group method can have several advantages over traditional interviews. When using focus groups, participants' interactions with each other can yield insights that would not be possible to obtain with individual interviews (Kitzinger, 1995 ). In addition to being able to complement each other, participants have the opportunity to respond to each other's answers, making it easier to identify differences between their views. These differences can further be used to clarify the reasons behind participants' views (Kitzinger, 1994 ). Finally, with regard to social desirability, research has also found that focus groups, when compared to individual interviews, can actually induce participants to take a more critical stance (Watts and Ebbutt, 1987 ; Kitzinger, 1995 ). What matters here is to create a safe atmosphere for participants in which to express their views (Kitzinger, 1995 ).

For this study, we chose to focus on effective self-regulators, rather than making a comparison between effective vs. less effective students. Rather than focusing on the factors that influence effective self-regulation and the incorporation of desirable difficulties, the aim of this study was to take a step back and come to a comprehensive picture of what this effectiveness actually looks like.

In summary, in order to acquire more in-depth insight into the variation of students' strategy use and the reasons behind it, these considerations led us to choose a focus group approach to study students' self-regulation and incorporation of desirable difficulties into their learning. We complemented the focus group approach with a traditional learning strategy survey (cf., Hartwig and Dunlosky, 2012 ) to compare and contrast results between approaches and analyze the value of each. The research questions guiding this study were: How do highly effective self-regulating students in a PBL higher-education curriculum approach their learning? How do they incorporate desirable difficulties into this process?

This study took place in the context of the first and second year of the 6-year undergraduate medical program at Maastricht University. This university uses a problem-based learning (PBL) format, in which learning takes place starting from authentic, real-world cases (Schmidt, 1983 ). Students work on these problems in small tutorial groups, typically consisting of approximately 10–12 students. These tutorial sessions are moderated by a tutor, who is expected to act as a facilitator, rather than as a knowledge transmitter. To structure the PBL process, Maastricht University uses a seven-step model called the Seven-Jump (Moust et al., 2005 ), consisting of clarification of terms, problem definition, brainstorming about possible explanations to the problem, structuring and analysis of the identified explanations, identification of learning questions, self-study, and post-discussion aimed at integrating individual students' findings. The first five steps take place in one tutorial group session, after which students individually study the literature to answer the learning questions outside the tutorial group. A few days later, the tutorial group gets together again to discuss their findings in the post-discussion, after which the cycle repeats for a new problem. In this curriculum, the academic year is divided into six courses, ranging between four to 8 weeks, each focusing on a specific multidisciplinary topic. At the end of a course, students are tested with an exam focused on the contents of this course (mostly multiple-choice).

Given its emphasis on students' independent literature search and self-study, the PBL format provides a fruitful context for the study of students' use of learning strategies and incorporation of desirable difficulties. Specifically, as students are required to find their own literature and use it to independently answer their learning questions they will need a range of strategies to manage this process and monitor their understanding, leading to a large pool of potential strategies for students to report on. This situation offers a unique potential to gain insight into what constitutes an effective approach.

Participants

In order to come to a picture of effective strategy use and the incorporation of desirable difficulties for students in a PBL curriculum, we used a purposive sampling strategy (Ritchie et al., 2013 ). At the end of the first year (academic year 2013–2014), mentors of first-year undergraduate medical students were asked to identify students who they perceived to use effective learning strategies (the instructions for the mentors can be found in Appendix A in the Supplementary Material). Sixteen mentors identified 42 students for the study. These students were approached by e-mail to invite them for our study, to be held at the beginning of their second year (academic year 2014–2015). Thirty students (71%) indicated willingness to participate. Two students indicated it would not be possible to be present at the times the focus groups were held. Two students filled out the learning strategy questionnaire (see below) but did not attend the focus groups, and were therefore excluded from further analysis. The final number of students participating in the focus groups was therefore N = 26, of which 20 students were female, ages ranging between 18 and 23 years old (one student did not provide an age). The total number of students enrolled for the tutorial groups at the beginning of Year 2 was 298. Written informed consent was obtained from all participants prior to the start of the study. The study was approved by the ethical review board of the Netherlands Association for Medical Education (file number 402). Students were offered a small monetary gift voucher as a reward for their participation in the study.

Learning Strategy Questionnaire

At least one week prior to the focus groups, students were asked to fill out a learning strategy questionnaire. We adapted the questionnaire used by Hartwig and Dunlosky ( 2012 ) to fit our PBL learning situation. Specifically, we adapted the wording of the questionnaire to refer to the tutorial group meetings that students encounter in the PBL setting. Furthermore, rather than asking students whether they do or do not use a specific strategy regularly (using a binary yes/no format), we used a Likert scale asking students how often they use these strategies while studying, ranging from 1 (never) to 5 (every study session). This was also applied to the question of whether students go back to course material after a course has ended, and whether students read study sources more than once. For the questions asking students on what parts of the day they study most and on what parts of the day they study most effectively, “evening” and “late night” were combined into one category (“evening”). Furthermore, the strategy questions were adapted to reflect the ones most relevant for the current educational context.

We dropped the question asking students whether they study more for open questions or multiple-choice questions, as the tests that medical students encounter in the program are mostly multiple-choice. The question of how students decide what to study next was posed as an open question. Finally, in order to reflect the focus of our study, we added four questions: (1) How did you develop the study strategies you are using now (open question, replacing the question if whether students' study strategies were taught to them by a teacher), (2) If you had the time and somebody would explain it to you, would you want to change your study strategies (yes/no), (3) What and why would you then want to change (open question), and (4) What kind of education would you most appreciate to change your study strategies? Think about: lectures, videos, practice with a trainer, etc. (open question). Finally, we added a question asking students for any further comments they may have. All questions not rated on a Likert scale (open questions and study times) were thematically coded by two raters. Inconsistencies were discussed until consensus was reached.

With this questionnaire we attempted to obtain a baseline measure of students' strategy use ( what are the strategies that are used), to later complement this with the in-depth focus groups ( how are the strategies used). In summary, the adapted questionnaire consisted of 10 questions assessing students' strategy use, using a Likert scale ranging from 1 (never) to 5 (every study session), as well as one question allowing students to list other strategies they use during studying. Furthermore, there were 12 questions inquiring about additional aspects of students' study behavior, for example, preferred study time (with five questions being open ended). Appendix B in the Supplementary Material provides an overview of the questionnaire.

Focus Groups

Students were divided into four separate focus groups. Each focus group lasted ~1 to 1.5 h. Each focus group was moderated by the second author and observed by the last author and a student assistant. The second author is an educational scientist by background and specializes in qualitative methodology. The last author specializes in effective study strategies. She served as an observer, in order to avoid influencing the results or “leading” the participants. The student assistant observed as well and organized the focus groups. Based on a vignette approach, students were asked how they prepare for different educational activities in the PBL medical curriculum. A total of six vignettes was used (see Appendix C in the Supplementary Material for the interview protocol, including the vignettes used). These vignettes concerned the post-discussion, exam, progress test, skills lab, Pscribe (written assignments assessing students' pharmacotherapeutic reasoning) and extracurricular activities. To answer our research question related to students' learning strategies during self-study, we focused our analysis on the first two vignettes (post-discussion and exam). After 4 months, students were invited back for a second focus group meeting, in which we discussed preliminary results, in order to check our interpretation of the findings (member checking), and to see whether students were consistent in their reports. Two students did not attend the second meeting because the interview dates did not fit their schedule.

The interview protocol used for the focus groups can be found in Appendix C in the Supplementary Material.

All focus groups were audio recorded and transcribed verbatim. A constructivist grounded theory methodology (Charmaz, 2014 ) was taken when analyzing the data. In grounded theory, the aim is to generate a theory or understanding of a certain process (Creswell, 2007 ). In a process of iterative data analysis, the researchers go through the different steps of open coding (generating initial codes for data categories), axial coding (identifying a core phenomenon and its surrounding categories), and selective coding (connecting categories and developing the theory). We chose this approach due to our focus on understanding the process of effective strategy use and incorporation of desirable difficulties, with a strong interest in the conditions that support or hinder this process (Creswell, 2007 ).

Initial, open coding was done by the first author. This was done in a line-by-line fashion, in which representative codes were assigned to the participants' utterances. During this process, several meetings were held with the second and last author to discuss the codes. After arriving at an initial codebook, codes were related to each other in a process of axial coding. During this process, codes were compared and contrasted with each other, looking for connections in order to create themes from overlapping codes. This step was initially done by the first author, with the second and last author each coding a non-overlapping 25% of the codebook to ensure rigor. Findings from this step were discussed until consensus was reached. Results from the analysis were discussed with the third and fourth author. Finally, in a process of selective coding by the first, second and last author, themes were related to each other in order to come to an overarching model of the data.

Tables ​ Tables1 1 , ​ ,2 2 show the results from the survey on students' strategy use and the additional aspects of students' study behavior, respectively.

Means and standard deviations for students' responses on the learning strategy questions, from highest to lowest mean.

Summary of students' responses to questions about additional aspects of their study behavior.

Interestingly, the students in our sample indicate a frequent use of the strategies regarded in the literature as effective, such as self-testing, questioning and self-explanation (Hartwig and Dunlosky, 2012 ; Dunlosky et al., 2013 ), indicating that our purposeful sampling strategy was effective. Furthermore, as indicated in Table ​ Table2, 2 , students report spacing their tutorial preparations over multiple sessions, indicating use of distributed practice (Dunlosky et al., 2013 ). However, as indicated by Table ​ Table2, 2 , students also report using some of the strategies that are typically viewed as ineffective for reaching long-term retention and transfer, particularly summarizing, mental imagery and underlining/marking (Dunlosky et al., 2013 ).

When responding to the question about which other strategies they use, strategies students reported (restricted to the ones not covered by the questionnaire) were: preparing their case on their laptop and shortly summarizing it before the tutorial group, writing out practical activities and going over this information during the exam week, drawing or writing out difficult things, making practice tests and correcting incorrectly answered items, watching videos, making diagrams after studying a case to summarize as much as possible, making concrete and compact cases, working in a disciplined manner, creating mind maps and drawings, drawing figures or pictures, rereading summaries, writing down and rereading difficult parts, printing out all cases and information from practicals and putting them together in one-folder to create an overview of the entire course, rehearsing lectures, and attentively working out learning materials in the case.

When responding to the question asking students whether they had any further comments, students emphasized the importance of lectures, active processing of learning materials through the creation of summaries, the added value of PBL and discussions during tutorial groups, and the importance of keeping order in the learning materials to avoid missing information.

In the focus groups, students were asked about their study approaches, in order to gain more insight into the ways in which they use their learning strategies.

Using the constructivist grounded theory methodology, a model was constructed describing how highly effective strategy users approach their learning. The results of this process are depicted in Figure ​ Figure1. 1 . In this model, students are driven by a personal learning goal, adopting either a qualitative or quantitative approach to learning. When learning, these highly effective strategy users are continually engaged in active processing of subject matter, while monitoring their understanding of the content and adjusting their approach when necessary. This process is guided by a constant balance between adhering to established study habits, while maintaining a sufficient degree of flexibility to adapt to changes in the learning environment, assessment demands and time limitations. Although students demonstrated metacognitive knowledge of the effectiveness of their strategies and the reasons for using them, this was not the case for all aspects of their strategy use. Indeed, students reported using several strategies which are traditionally regarded as “ineffective” (highlighting, rereading etc.), but used them in a way that helped them adjust to their learning situation and goal.

Model describing highly effective strategy users' approach to learning.

In the following, we will describe the different components of this model, the implications for students' self-regulation, and the incorporation of desirable difficulties into their learning.

Quantity and Quality

During the focus groups, many of the students described being driven by a personal learning goal, adopting a quantitative or qualitative approach to learning. Specifically, quantitatively oriented students used numerical indicators as the basis for their learning. For example, when referring to collecting information for the post-discussion, one student stated:

“ And then, yeah, just translate it a little bit and write it down in my own words. And uh, yeah, then I just have about fifteen pages usually. And when I have three pages I really feel like I, yeah, have too little .” Focus group 1, session 1, participant A

On the other hand, students adopting a qualitative approach emphasized the quality of their materials and of their understanding. Rather than focusing on how much material they had produced, these students would focus on how well they understood and remembered what they had studied. As one student explained:

“ Well, if you have 7 pages and you don't understand any of it, you haven't achieved anything in the end. You'll have a lot of material to study and when you study you can brag about having a 50-page case .” Focus group 1, session 1, participant B

Active Processing and Monitoring

During focus group discussions it became clear that students were continually engaged in active processing of the subject matter, while monitoring their understanding of the content and adjusting their approach when necessary. In this sense, students are incorporating desirable difficulties into their learning, as they are not content with passively reading the subject matter, but try to find ways to be actively engaged.

“ You should never literally copy an entire text. Or [you should do it] in the way he [other participant] does it, explain it or write it in your own words, but do something that makes it your own.” Focus group 4, session 1, participant A

In some cases, the PBL system at Maastricht University was indicated as a contributing factor to this active approach, as students are required to be able to discuss their findings in the post-discussions. This became clear in the words used to describe it:

“ I think that is really the key, treat the subject matter in an active way. You're in Maastricht, this is what they ask from you and it also just works.” Focus group 1, session 1, participant C

“ Well I had, yes in [a different city] I really had to learn from books. (…), so I think that that is just, that's not possible here, in Maastricht you also have to be able to tell everything coherently. So then I made a mix from that, that I, because I was good at studying from books, but that I could also reproduce it in the tutorial group.” Focus group 2, session 1, participant A

In addition to this active processing, students reported a continuous monitoring of understanding, and adjusting their learning when necessary. In many cases, this monitoring was achieved by various forms of self-testing. A commonly reported tactic for this was explaining the subject matter to another person, either physically or hypothetically:

“ (…) sometimes it is nice when people are like asking questions. Then I hear myself explaining it and then I hear whether I understand it, so to speak.” Focus group 3, session 1, participant A

“ And, uhm, when I look through my case at the end I should actually be able to explain each component that I discuss to someone else. I don't actually do that, but I should be able to.” Focus group 1, session 1, participant B

Also, students often used externally provided resources such as practice tests to test their knowledge and understanding. Interestingly, the strategies students reported using to correct learning when this monitoring revealed knowledge deficits, were mostly surface-level strategies such as rereading. However, self-testing is an important strategy to improve learning (Roediger and Karpicke, 2006 ) and has to be actively built into the learning process. The fact that students reported using practice tests and testing themselves indicated again students' willingness to incorporate desirable difficulties into their learning.

Habits and Flexibility

Students' learning process, as guided by their learning goal and characterized by active processing of subject matter and continuous monitoring of understanding, is further guided by a constant balance between adhering to established study habits, while maintaining a sufficient degree of flexibility to adapt to changes in the learning environment, assessment demands and time limitations. For example, students often indicated they had fixed times or places for studying, or a fixed order in which to process the materials for studying. At the same time, students also found themselves in situations where they had to adapt to changes in their learning situation and reported several strategies to maintain this flexibility. For example, one student indicated photocopying book sections in advance to be able to study when going home to parents during the weekends, thereby maintaining flexibility in time and location on which to study. This flexibility was also evident in students' strategy use. Students reported that they had experimented with strategies over the years, finding out “what works for them.” While some students indicated that they had not experienced the need to change their strategies, because they felt comfortable with their strategies and were happy with the results they produced, others indicated that they had used criteria such as their performance as benchmarks for whether or not they should adjust their strategies. As one student indicated:

“ I think this is something in which you are supposed to grow and if you keep telling yourself that your own strategy works and you score 6's [points out of 10, 10 being the highest] then you're actually doing something wrong. But then you're just, I would almost say lazy, you just don't feel like changing it .” Focus group 1, session 1, participant B

Furthermore, several students indicated adapting their strategies according to the demands of the test. Although some students reported using the same studying methods regardless of the way of questioning on the test, others indicated adapting their strategies depending on whether they would have to answer multiple choice questions (focusing on retention and recognition) or open questions (studying more, with a stronger focus on understanding).

Metacognitive Knowledge

Although students demonstrated metacognitive knowledge of the effectiveness of their strategies and the reasons for using them, this was not the case for all aspects of their strategy use. Students indicated in the second session that, when given a list of all strategies mentioned in the first session and asked to indicate which strategies they used most often, it was difficult to label these strategies properly. As one student indicated:

“ (…) with me mostly with visualizing, that I didn't realize that I was doing it or how I was doing it, until I wrote down that I was doing it. Then I thought, oh yes, I do this quite a lot.” Focus group 1, session 2, participant B

It was especially difficult for students to indicate how they monitored their understanding, or how they distinguished between important and less important topics and how deep to process the information. Many students indicated this was a “feeling,” or something they had learned from experience.

Furthermore, students reported using several strategies which are traditionally regarded as “ineffective,” such as highlighting and rereading of text (Dunlosky et al., 2013 ). However, students used them in a way that helped them adjust to their learning situation, by using the strategies in an active way. Although there were some exceptions (e.g., highlighting text in order to reread it afterwards), examples include repeating subject matter using different sources and media, making handwritten summaries to be actively engaged with the subject matter, paraphrasing in order to monitor understanding, or rereading text to check whether it still makes sense in the context of clinical practice.

In fact, in one of the focus groups students indicated the need to incorporate desirable difficulties into their learning process, emphasizing the wish to attain long-term retention, rather than short-term storage, in order to become a competent doctor after graduation. Students often recognized the need to invest effort in learning, as opposed to relying on low-effort surface-level strategies (for example, purposefully using English rather than Dutch books, as the additional effort required prevents a shallow reading of the text). An overarching theme in this regard was a focus on creating understanding, finding the logic in the educational material and making connections between different topics and educational activities, as opposed to for example rote learning or memorizing symptoms. One student explained:

“ I always do that [check if you can apply the case to medical practice] , I always try to make the case explainable. Just because I like that, then I know that I understand and when it is written down on sheets everywhere then I think oh, why is this value high or that value low. Or, because that lab test, oh yes, that makes sense too. It is not that I will think about what it is [come up with a diagnosis], but I do check to see if it makes sense or not” Focus group 2, session 1, participant B

In summary, the participants in our study use a variety of strategies to regulate their learning and to incorporate desirable difficulties into this process. In addition to active processing of subject matter and a continuous monitoring of understanding, participants understand the need to obtain long-term storage and understanding, rather than short-term results, in some cases prompted by the perspective of having to become a capable doctor.

This paper outlines the results of a study investigating highly effective strategy users' approaches to learning. As a starting point, a survey was administered to students asking about how their study strategies and how they approach their learning. Results from this survey indicated students' adherence to some highly effective strategies (e.g., self-testing), but also the use of some of the less effective strategies (e.g., highlighting). Afterwards, focus groups were organized in order to gain insight into how students use these learning strategies. Specifically, as survey data can provide insight into which strategies students use and how often they use them, the qualitative approach can clarify why students use these strategies, under which circumstances, and how flexible they are regarding their use.

Based on the focus groups, a model was constructed which describes how these students prepare for different learning activities. The first element in our model, as emanating from the focus groups was the distinction between quantitatively vs. qualitatively oriented students. The students who mentioned having a learning goal, expressed this in a way that suggests a sharp distinction between these two opposites: students are either quantitatively or qualitatively oriented. However, from a motivational or self-regulatory perspective, one would expect this variable to fall along a continuum (Ryan and Deci, 2000 ), with students leaning more toward either side of the spectrum depending on varying contexts and conditions. For example, it is conceivable that students who have a predominantly qualitative orientation might become more quantitatively oriented in the face of insecurities or time constraints. Conversely, generally quantitatively oriented students might adopt a more qualitative orientation when studying topics they are highly interested in. Possibly, students who did not express a learning goal might fall somewhere along this spectrum (a point we have tried to emphasize by adding the dotted line connecting the two opposites). Validating the polarized vs. continuous nature of this distinction, as well as determining the factors that influence students' respective orientations, could be an interesting avenue for future research.

The second theme concerned students maintaining a continuous balance between established habits vs. a flexibility to meet changing demands. Indeed, this would make sense from a desirable difficulties perspective, as these students do not “give up” in the face of changing demands, but rather persist and adapt to the situation in order to reach their goals. Earlier research also correlated flexibility (termed adaptive control) with self-regulated learning, deep processing, and a propensity to undertake effortful cognitive activities (Evans et al., 2003 ). In terms of implications, several follow-up questions can be asked. First, what is the optimal combination between habits and flexibility? Will this balance be different in less effective students? What are students' core habits? What should be flexible, and what should be stable? What can be taught? Interventions should focus on optimizing this balance. Monitoring of understanding could be at the core of these interventions. When students have an accurate insight into which aspects they do and do not understand, and which strategies lead to a better understanding, it can be easier to make decisions about which strategies need to remain stable, and which should be adapted.

The third theme arising from the data, which characterized students' learning process, was students' continuous engagement in active processing of the learning material and monitoring of understanding. In addition to being aspects of effective (self-regulated) strategy use (Zimmerman, 1990 ; Dunlosky et al., 2013 ), it is also possible that this result can (at least partly) be attributed to the PBL curriculum in which this study took place, as these learning methods are hallmarks of this instructional approach (Hmelo-Silver, 2004 ; Loyens et al., 2008 ). Indeed, one of the students in the focus group even indicated the problem-based curriculum as a reason for adopting an active approach to learning. Given the fact that this study has only been carried out in a PBL context, it is difficult to disentangle these influences. Future studies could seek to unravel these factors further.

The final theme emerging from the focus groups concerned students' metacognitive knowledge. Interestingly, students reported using several strategies which traditional self-report questionnaires tend to treat as “ineffective,” but used them in an active way to help cope with the demands of their specific learning situation. This indicates that what matters most is not which strategies students use, but rather how they use them. In other words, students adapted strategies to fit their particular learning situation. Indeed, students' adaptability in their strategy use has been identified by other authors as an important feature of effective self-regulation in students (Broekkamp and Van Hout-Wolters, 2007 ). This sense of flexibility was also evident in other parts of the model, where students maintained a continuous balance between established study habits on the one hand, and a sense of flexibility to deal with changes on the other.

Another reason for students' use of surface-level learning strategies could be the form of assessment. Students are often assessed with multiple-choice question tests or open question tests focused solely on short-term retention of information. Several studies have found that students will adapt their strategies based on what they perceive will be expected of them on the examination (Thomas and Rohwer, 1986 ; Broekkamp and Van Hout-Wolters, 2007 ). Indeed, students in our study indicated changing their strategies according to whether questions would be asked in a multiple choice vs. an open question format. In this sense, rather than being “ineffective,” these surface-level strategies could be interpreted as being highly efficient in terms of the (short-term) goal students are aiming to achieve, if this goal is to obtain a good grade on the retention-based exam (Morris et al., 1977 ). If the goal of the curriculum is for students to strive for deep-level processing and understanding, the test demands need to be aligned with this objective (Broekkamp and Van Hout-Wolters, 2007 ), asking questions that will require this approach from students.

On the other hand, several students indicated an understanding of the need to obtain long-term retention and understanding, an inclination that seemed to be promoted by a desire to become a capable doctor. This can have important implications for interventions aimed at improving self-regulation for students who are less skilled self-regulators. Specifically, if interventions would focus on aiding students in attaining a clear perspective of their goals and long-term profession, this could improve their self-regulatory behavior and intention to build in desirable difficulties into their learning. Although we did not originally set out to investigate the link between students' learning behavior and their future time perspective, previous work has been done to establish this link, with research indicating that students' long-term time perspectives are associated with adaptive self-regulatory strategies and deep cognitive processing (Bembenutty and Karabenick, 2004 ; de Bilde et al., 2011 ). As these studies are mostly correlational, the direction of these effects is not entirely clear. Future research could try to establish the direction of causality by employing a longitudinal (de Bilde et al., 2011 ) or experimental approach.

The model identified can elaborate on existing theoretical models of metacognition by explicating the criteria students use to monitor and control their learning and how they adapt their strategies to fit their learning needs. For example, Nelson and Narens ( 1990 ) outline a theoretical framework in which students' allocation of study time is determined by their judgments about the difficulty it takes them to master certain information (ease of learning; EOL), their judgments about how well they have mastered certain recallable information (judgments of learning; JOL), and the degree to which they believe they have previously known currently unrecalled information (feeling of knowing; FOK). Their research found that students will allocate extra study time based on their EOL, JOL, and FOK judgments, with students studying general information items generally allocating extra study time to information with a lower EOL (meaning they are judged to be harder), higher FOK, and lower JOL. Also when it comes to the allocation of restudy , students will allocate this restudy time to information they judge as poorly learned (Nelson et al., 1994 ). The current study adds to this literature by shedding light on some of the criteria students may use to make these judgments. Specifically, students seem to focus on qualitative or quantitative criteria for making these judgments. Furthermore, for FOK, Nelson and Narens ( 1990 ) indicate that these judgments monitor the recallable aspects of the information a student has in memory (such as whether they have used it to correctly answer a question before). This could potentially explain the differences between the qualitative and quantitative orientations found in our study. For some students, the qualitative aspects related to the studied information may be hard to recall. For example, some of the information may never have been tested yet, making it difficult for students to derive these judgments. This may lead them to focus on more easily recallable, quantitative information instead.

Following this line of reasoning, this focus on easily recallable, quantitative aspects of learning may lead students to adopt more surface-level strategies, as these might be sufficient to satisfy the quantitative criteria. Indeed, Koriat ( 1997 ) found that extrinsic cues are less informative for students' JOLs than intrinsic cues, and these inaccurate JOLs could in turn lead to inadequate study strategies. Although students in our study seemed to follow the same general path of self-regulation, the qualitative approach might lead to more elaborative learning strategies and incorporation of desirable difficulties. However, a focus on quantitative criteria is apparently sufficient for students to pass their exams and be successful in university (a point which was already elaborated upon above). However, we do not have any information about their long-term retention. Future studies should focus on more elaborative learning outcomes and longer retention intervals, to further unravel the potentially differential effects of the different approaches to learning.

This study has several limitations. First, our focus groups were limited to second-year undergraduate medical students who were effectively self-regulating their learning. Given the PBL context in which these students are learning, this provided a fruitful basis to start from when investigating effective students' approaches to learning, but we cannot be sure about how these findings relate to other student populations. Furthermore, our study was limited to students from the undergraduate medical program. It is possible that there are characteristics in this program, which are not easily transferable to programs focusing on other domains. A specific example of this can be found in the long-term perspective that several students indicated as the basis for their desire to understand the subject matter, as hinted at above. In a study program like Medicine, the end goal of becoming a doctor is quite clear. In many other undergraduate programs, this long-term perspective may be less evident. Future research could look into what constitutes effective self-regulation in other study programs and other, non-PBL oriented universities. Furthermore, although the purpose of this study was to illustrate effective self-regulation rather than to contrast different groups of students, it would be interesting to see what picture will emerge when asking the same questions to low self-regulating students. We have tried to ensure replicability by providing rich descriptions of context, methods, and results, in an attempt to increase opportunities for judgments of transferability.

Related to the distinction between effective vs. ineffective strategy users is the questions of whether we were able to correctly identify which students were effective strategy users. We used students' mentors as informants for our purposeful sampling strategy. We have confidence in this strategy, as mentors are among the few key persons who have a bird's eye view of students' overall performance, for both the entire duration of the program, as well as in comparison to other students. They also discuss students' learning strategies at least two times during the first year in an individual mentor meeting. However, their judgments are inherently subjective, and although they were given instructions on what is meant by effective strategy users, we have no insight into their decision making when they selected these students. Although it was a conscious decision not to include grades as a measure of self-regulation (as students using shallow strategies may very well obtain good test results in the short term), it could be worthwhile to think about other ways to triangulate students' strategy effectiveness.

Finally, we chose to use learning questionnaire used by Hartwig and Dunlosky ( 2012 ) as a starting point for our study, in order to build further on this work and demonstrate the added value of the focus groups in this context. However, as this survey measures each strategy by only one item, it was not possible to compute reliability or internal consistency estimates. This problem is mitigated by the fact that we used the survey as a starting point for our focus groups, rather than conducting analyses analyzing differences between groups or as a result of some intervention. However, the research design could be strengthened by adding more items per strategy, in order to be able to make inferences about the reliability and internal consistency of students' responses.

Overall, this study contributes to the literature by providing an in-depth, qualitative description of how highly self-regulated medical students in a PBL curriculum approach their learning and build in desirable difficulties in their learning process. This model can serve as a framework for further study into the various factors that influence (effective) self-regulation, and as a starting point for designing interventions focused on improving strategy use in less effective students.

Author Contributions

AdB and RS were responsible for the design and data collection of the study. SR performed analysis of the data, in close collaboration with AdB and RS. SR drafted the article, incorporating edits, and feedback from all other authors (AdB, RS, JvM, and HS). All authors made a substantial contribution to the interpretation of the data for this work.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding. This research was funded by the Netherlands Organization for Scientific Research (Veni grant number 451-10-035).

Supplementary Material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2018.02501/full#supplementary-material

• Arbesman S. (2013). The Half-life of Facts. Why Everything We Know has an Expiration Date . New York, NY: Penguin; Group. [ Google Scholar ]
• Bembenutty H., Karabenick S. A. (2004). Inherent association between academic delay of gratification, future time perspective, and self-regulated learning . Educ. Psychol. Rev. 16 , 35–57. 10.1023/B:EDPR.0000012344.34008.5c [ CrossRef ] [ Google Scholar ]
• Bernacki M. L., Byrnes J. P., Cromley J. G. (2012). The effects of achievement goals and self-regulated learning behaviors on reading comprehension in technology-enhanced learning environments . Contemp. Educ. Psychol. 37 , 148–161. 10.1016/j.cedpsych.2011.12.001 [ CrossRef ] [ Google Scholar ]
• Bjork R. A. (1994). Memory and metamemory considerations in the training of human beings , in Metacognition: Knowing About Knowing , eds Metcalfe J., Shimamura A. (Cambridge, MA: MIT Press; ), 185–205. [ Google Scholar ]
• Blasiman R. N., Dunlosky J., Rawson K. A. (2017). The what, how much, and when of study strategies: comparing intended versus actual study behaviour . Memory 25 , 784–792. 10.1080/09658211.2016.1221974 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Boud D., Falchikov N. (2006). Aligning assessment with long-term learning . Assess Eval High Edu. 31 , 399–413. 10.1080/02602930600679050 [ CrossRef ] [ Google Scholar ]
• Bråten I., Samuelstuen M. S. (2007). Measuring strategic processing: comparing task-specific self-reports to traces . Metacogn. Learn. 2 , 1–20. 10.1007/s11409-007-9004-y [ CrossRef ] [ Google Scholar ]
• Broekkamp H., Van Hout-Wolters B. H. (2007). Students' adaptation of study strategies when preparing for classroom tests . Educ. Psychol. Rev. 19 :401 10.1007/s10648-006-9025-0 [ CrossRef ] [ Google Scholar ]
• Charmaz K. (2014). Constructing Grounded Theory. Thousand Oaks, CA: Sage. [ Google Scholar ]
• Creswell J. W. (2007). Qualitative Inquiry and Research Design: Choosing Among the Five Approaches . Thousand Oaks, CA: Sage; Publications. [ Google Scholar ]
• de Bilde J., Vansteenkiste M., Lens W. (2011). Understanding the association between future time perspective and self-regulated learning through the lens of self-determination theory . Learn. Instruct. 21 , 332–344. 10.1016/j.learninstruc.2010.03.002 [ CrossRef ] [ Google Scholar ]
• Dunlosky J., Rawson K. A., Marsh E. J., Nathan M. J., Willingham D. T. (2013). Improving students' learning with effective learning techniques: Promising directions from cognitive and educational psychology . Psychol. Sci. Public Inter. 14 , 4–58. 10.1177/1529100612453266 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Evans C. J., Kirby J. R., Fabrigar L. R. (2003). Approaches to learning, need for cognition, and strategic flexibility among university students . Br. J. Edu. Psychol. 73 , 507–528. 10.1348/000709903322591217 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Hartwig M. K., Dunlosky J. (2012). Study strategies of college students: are self-testing and scheduling related to achievement? Psychon. Bull. Rev. 19 , 126–134. 10.3758/s13423-011-0181-y [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Helsdingen A., Van Gog T., Van Merriënboer J. (2011). The effects of practice schedule and critical thinking prompts on learning and transfer of a complex judgment task . J. Educ. Psychol. 103 :383 10.1037/a0022370 [ CrossRef ] [ Google Scholar ]
• Hmelo-Silver C. E. (2004). Problem-based learning: What and how do students learn? Educ. Psychol. Rev. 16 , 235–266. 10.1023/B:EDPR.0000034022.16470.f3 [ CrossRef ] [ Google Scholar ]
• Karpicke J. D., Butler A. C., Roediger Iii H. L. (2009). Metacognitive strategies in student learning: do students practise retrieval when they study on their own? Memory 17 , 471–479. 10.1080/09658210802647009 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Kirschner P. A., Sweller J., Clark R. E. (2006). Why minimal guidance during instruction does not work: an analysis of the failure of constructivist, discovery, problem-based, experiential, and inquiry-based teaching . Educ. Psychol. 41 , 75–86. 10.1207/s15326985ep4102_1 [ CrossRef ] [ Google Scholar ]
• Kitzinger J. (1994). The methodology of focus groups: the importance of interaction between research participants . Sociol. Health Illness 16 , 103–121. 10.1111/1467-9566.ep11347023 [ CrossRef ] [ Google Scholar ]
• Kitzinger J. (1995). Qualitative research. Introducing focus groups . BMJ 311 :299 10.1136/bmj.311.7000.299 [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Koriat A. (1997). Monitoring one's own knowledge during study: a cue-utilization approach to judgments of learning . J. Exp. Psychol. 126 :349 10.1037/0096-3445.126.4.349 [ CrossRef ] [ Google Scholar ]
• Kornell N., Bjork R. A. (2007). The promise and perils of self-regulated study . Psychon. Bull. Rev. 14 , 219–224. 10.3758/BF03194055 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Loyens S. M. M., Magda J., Rikers R. M. J. P. (2008). Self-directed learning in problem-based learning and its relationships with self-regulated learning . Educ. Psychol. Rev. 20 , 411–427. 10.1007/s10648-008-9082-7 [ CrossRef ] [ Google Scholar ]
• McCardle L., Hadwin A. F. (2015). Using multiple, contextualized data sources to measure learners' perceptions of their self-regulated learning . Metacogn. Learn. 10 , 43–75. 10.1007/s11409-014-9132-0 [ CrossRef ] [ Google Scholar ]
• Morris C. D., Bransford J. D., Franks J. J. (1977). Levels of processing versus transfer appropriate processing . J. Verbal Learning Verbal Behav. 16 , 519–533. 10.1016/S0022-5371(77)80016-9 [ CrossRef ] [ Google Scholar ]
• Moust J. H. C., Berkel H. J. M. V., Schmidt H. G. (2005). Signs of erosion: reflections on three decades of problem-based learning at maastricht university . High. Edu. 50 , 665–683. 10.1007/s10734-004-6371-z [ CrossRef ] [ Google Scholar ]
• Nelson T. O., Dunlosky J., Graf A., Narens L. (1994). Utilization of metacognitive judgments in the allocation of study during multitrial learning . Psychol. Sci. 5 , 207–213. 10.1111/j.1467-9280.1994.tb00502.x [ CrossRef ] [ Google Scholar ]
• Nelson T. O., Narens L. (1990). Metamemory: a theoretical framework and new findings , in Psychology of Learning and Motivation , Vol. 26 , ed Bower G. H. (New York, NY: Academic Press; ), 125–173. [ Google Scholar ]
• Panadero E., Klug J., Järvelä S. (2016). Third wave of measurement in the self-regulated learning field: when measurement and intervention come hand in hand . Scand. J. Edu. Res. 60 , 723–735. 10.1080/00313831.2015.1066436 [ CrossRef ] [ Google Scholar ]
• Perry N. E., Winne P. H. (2006). Learning from learning kits: gStudy traces of students' self-regulated engagements with computerized content . Educ. Psychol. Rev. 18 , 211–228. 10.1007/s10648-006-9014-3 [ CrossRef ] [ Google Scholar ]
• Ritchie J., Lewis J., Nicholls C. M., Ormston R. (2013). Qualitative Research Practice: A Guide for Social Science Students and Researchers . Thousand Oaks, CA: Sage. [ Google Scholar ]
• Roediger H. L., Karpicke J. D. (2006). Test-enhanced learning: taking memory tests improves long-term retention . Psychol. Sci. 17 , 249–255. 10.1111/j.1467-9280.2006.01693.x [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Ryan R. M., Deci E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being . Am Psychol . 55 , 68–78. 10.1037/0003-066X.55.1.68 [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Schellings G. (2011). Applying learning strategy questionnaires: problems and possibilities . Metacogn. Learn. 6 , 91–109. 10.1007/s11409-011-9069-5 [ CrossRef ] [ Google Scholar ]
• Schmidt H. G. (1983). Problem based learning: rationale and description . Med. Educ. 17 , 11–16. 10.1111/j.1365-2923.1983.tb01086.x [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Schunk D. H., Zimmerman B. J. (1994). Self-regulation of Learning and Performance: Issues and Educational Applications . Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. [ Google Scholar ]
• Thomas J. W., Rohwer W. D. (1986). Academic studying: the role of learning strategies . Educ. Psychol. 21 , 19–41. 10.1080/00461520.1986.9653022 [ CrossRef ] [ Google Scholar ]
• Van Merriënboer J. J., De Croock M. B., Jelsma O. (1997). The transfer paradox: effects of contextual interference on retention and transfer performance of a complex cognitive skill . Percept. Mot. Skills 84 , 784–786. 10.2466/pms.1997.84.3.784 [ CrossRef ] [ Google Scholar ]
• Van Merriënboer J. J., Kirschner P. A. (2018). Ten Steps to Complex Learning: A Systematic Approach to Four-Component Instructional Design . New York, NY: Routledge. [ Google Scholar ]
• Veenman M. V. (2011). Alternative assessment of strategy use with self-report instruments: a discussion . Metacogn. Learn. 6 , 205–211. 10.1007/s11409-011-9080-x [ CrossRef ] [ Google Scholar ]
• Watts M., Ebbutt D. (1987). More than the Sum of the Parts: research methods in group interviewing . Br. Educ. Res. J. 13 , 25–34. 10.1080/0141192870130103 [ CrossRef ] [ Google Scholar ]
• Winne P. H., Hadwin A. F. (1998). Studying as self-regulated learning , in Metacognition in Educational Theory and Practice , eds Hacker D. J., Dunlosky J., Graesser A. C. (Mahwah: Lawrence Erlbaum; ), 27–30. [ Google Scholar ]
• Zimmerman B. J. (1990). Self-regulated learning and academic achievement: an overview . Educ. Psychol. 25 , 3–17. 10.1207/s15326985ep2501_2 [ CrossRef ] [ Google Scholar ]

ORIGINAL RESEARCH article

Students' metacognitive knowledge of learning-strategy effectiveness and their recall of teachers' strategy instructions provisionally accepted.

• 1 School of Natural Sciences and Health, Tallinn University, Estonia
• 2 Tallinn University, Estonia

The final, formatted version of the article will be published soon.

This study aimed to investigate students' metacognitive knowledge and reported use of surface and deep learning strategies. It also explored the extent to which students recall their teachers' recommendations for learning strategies and the relationship between these recollections and students' knowledge and reported use of strategies. A scenario-based questionnaire was used to set a learning goal in the area of biology. Students' metacognitive knowledge was assessed through perceived effectiveness and reported use of learning strategies. Additionally, open-ended questions allowed students to recall and report recommendations given by their teachers. We used personcentered methods to explore whether different types of recollections were related to reported strategy use. Among students who recollected that their teachers have recommended deep learning strategies, it was typical to value deep strategies higher than surface strategies and report using deep strategies. Also, it was atypical among those students to value surface level strategies and not use deep strategies.

Keywords: metacognitive knowledge, Learning Strategies, Teacher instruction, Configural frequency analysis (CFA), Strategy effectiveness

Received: 04 Oct 2023; Accepted: 22 Apr 2024.

Copyright: © 2024 Olop, Granström and Kikas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

* Correspondence: Mx. Joosep Olop, School of Natural Sciences and Health, Tallinn University, Tallin, Estonia

People also looked at

• Open access
• Published: 17 April 2024

Deciphering the influence: academic stress and its role in shaping learning approaches among nursing students: a cross-sectional study

• Rawhia Salah Dogham 1 ,
• Heba Fakieh Mansy Ali 1 ,
• Asmaa Saber Ghaly 3 ,
• Nermine M. Elcokany 2 ,
• Mohamed Mahmoud Seweid 4 &
• Ayman Mohamed El-Ashry   ORCID: orcid.org/0000-0001-7718-4942 5  

BMC Nursing volume  23 , Article number:  249 ( 2024 ) Cite this article

203 Accesses

Metrics details

Nursing education presents unique challenges, including high levels of academic stress and varied learning approaches among students. Understanding the relationship between academic stress and learning approaches is crucial for enhancing nursing education effectiveness and student well-being.

This study aimed to investigate the prevalence of academic stress and its correlation with learning approaches among nursing students.

Design and Method

A cross-sectional descriptive correlation research design was employed. A convenient sample of 1010 nursing students participated, completing socio-demographic data, the Perceived Stress Scale (PSS), and the Revised Study Process Questionnaire (R-SPQ-2 F).

Most nursing students experienced moderate academic stress (56.3%) and exhibited moderate levels of deep learning approaches (55.0%). Stress from a lack of professional knowledge and skills negatively correlates with deep learning approaches (r = -0.392) and positively correlates with surface learning approaches (r = 0.365). Female students showed higher deep learning approach scores, while male students exhibited higher surface learning approach scores. Age, gender, educational level, and academic stress significantly influenced learning approaches.

Academic stress significantly impacts learning approaches among nursing students. Strategies addressing stressors and promoting healthy learning approaches are essential for enhancing nursing education and student well-being.

Nursing implication

Understanding academic stress’s impact on nursing students’ learning approaches enables tailored interventions. Recognizing stressors informs strategies for promoting adaptive coping, fostering deep learning, and creating supportive environments. Integrating stress management, mentorship, and counseling enhances student well-being and nursing education quality.

Peer Review reports

Introduction

Nursing education is a demanding field that requires students to acquire extensive knowledge and skills to provide competent and compassionate care. Nursing education curriculum involves high-stress environments that can significantly impact students’ learning approaches and academic performance [ 1 , 2 ]. Numerous studies have investigated learning approaches in nursing education, highlighting the importance of identifying individual students’ preferred approaches. The most studied learning approaches include deep, surface, and strategic approaches. Deep learning approaches involve students actively seeking meaning, making connections, and critically analyzing information. Surface learning approaches focus on memorization and reproducing information without a more profound understanding. Strategic learning approaches aim to achieve high grades by adopting specific strategies, such as memorization techniques or time management skills [ 3 , 4 , 5 ].

Nursing education stands out due to its focus on practical training, where the blend of academic and clinical coursework becomes a significant stressor for students, despite academic stress being shared among all university students [ 6 , 7 , 8 ]. Consequently, nursing students are recognized as prone to high-stress levels. Stress is the physiological and psychological response that occurs when a biological control system identifies a deviation between the desired (target) state and the actual state of a fitness-critical variable, whether that discrepancy arises internally or externally to the human [ 9 ]. Stress levels can vary from objective threats to subjective appraisals, making it a highly personalized response to circumstances. Failure to manage these demands leads to stress imbalance [ 10 ].

Nursing students face three primary stressors during their education: academic, clinical, and personal/social stress. Academic stress is caused by the fear of failure in exams, assessments, and training, as well as workload concerns [ 11 ]. Clinical stress, on the other hand, arises from work-related difficulties such as coping with death, fear of failure, and interpersonal dynamics within the organization. Personal and social stressors are caused by an imbalance between home and school, financial hardships, and other factors. Throughout their education, nursing students have to deal with heavy workloads, time constraints, clinical placements, and high academic expectations. Multiple studies have shown that nursing students experience higher stress levels compared to students in other fields [ 12 , 13 , 14 ].

Research has examined the relationship between academic stress and coping strategies among nursing students, but no studies focus specifically on the learning approach and academic stress. However, existing literature suggests that students interested in nursing tend to experience lower levels of academic stress [ 7 ]. Therefore, interest in nursing can lead to deep learning approaches, which promote a comprehensive understanding of the subject matter, allowing students to feel more confident and less overwhelmed by coursework and exams. Conversely, students employing surface learning approaches may experience higher stress levels due to the reliance on memorization [ 3 ].

Understanding the interplay between academic stress and learning approaches among nursing students is essential for designing effective educational interventions. Nursing educators can foster deep learning approaches by incorporating active learning strategies, critical thinking exercises, and reflection activities into the curriculum [ 15 ]. Creating supportive learning environments encouraging collaboration, self-care, and stress management techniques can help alleviate academic stress. Additionally, providing mentorship and counselling services tailored to nursing students’ unique challenges can contribute to their overall well-being and academic success [ 16 , 17 , 18 ].

Despite the scarcity of research focusing on the link between academic stress and learning methods in nursing students, it’s crucial to identify the unique stressors they encounter. The intensity of these stressors can be connected to the learning strategies employed by these students. Academic stress and learning approach are intertwined aspects of the student experience. While academic stress can influence learning approaches, the choice of learning approach can also impact the level of academic stress experienced. By understanding this relationship and implementing strategies to promote healthy learning approaches and manage academic stress, educators and institutions can foster an environment conducive to deep learning and student well-being.

Hence, this study aims to investigate the correlation between academic stress and learning approaches experienced by nursing students.

Study objectives

Assess the levels of academic stress among nursing students.

Assess the learning approaches among nursing students.

Identify the relationship between academic stress and learning approach among nursing students.

Identify the effect of academic stress and related factors on learning approach and among nursing students.

Materials and methods

Research design.

A cross-sectional descriptive correlation research design adhering to the STROBE guidelines was used for this study.

A research project was conducted at Alexandria Nursing College, situated in Egypt. The college adheres to the national standards for nursing education and functions under the jurisdiction of the Egyptian Ministry of Higher Education. Alexandria Nursing College comprises nine specialized nursing departments that offer various nursing specializations. These departments include Nursing Administration, Community Health Nursing, Gerontological Nursing, Medical-Surgical Nursing, Critical Care Nursing, Pediatric Nursing, Obstetric and Gynecological Nursing, Nursing Education, and Psychiatric Nursing and Mental Health. The credit hour system is the fundamental basis of both undergraduate and graduate programs. This framework guarantees a thorough evaluation of academic outcomes by providing an organized structure for tracking academic progress and conducting analyses.

Participants and sample size calculation

The researchers used the Epi Info 7 program to calculate the sample size. The calculations were based on specific parameters such as a population size of 9886 students for the academic year 2022–2023, an expected frequency of 50%, a maximum margin of error of 5%, and a confidence coefficient of 99.9%. Based on these parameters, the program indicated that a minimum sample size of 976 students was required. As a result, the researchers recruited a convenient sample of 1010 nursing students from different academic levels during the 2022–2023 academic year [ 19 ]. This sample size was larger than the minimum required, which could help to increase the accuracy and reliability of the study results. Participation in the study required enrollment in a nursing program and voluntary agreement to take part. The exclusion criteria included individuals with mental illnesses based on their response and those who failed to complete the questionnaires.

socio-demographic data that include students’ age, sex, educational level, hours of sleep at night, hours spent studying, and GPA from the previous semester.

Tool two: the perceived stress scale (PSS)

It was initially created by Sheu et al. (1997) to gauge the level and nature of stress perceived by nursing students attending Taiwanese universities [ 20 ]. It comprises 29 items rated on a 5-point Likert scale, where (0 = never, 1 = rarely, 2 = sometimes, 3 = reasonably often, and 4 = very often), with a total score ranging from 0 to 116. The cut-off points of levels of perceived stress scale according to score percentage were low < 33.33%, moderate 33.33–66.66%, and high more than 66.66%. Higher scores indicate higher stress levels. The items are categorized into six subscales reflecting different sources of stress. The first subscale assesses “stress stemming from lack of professional knowledge and skills” and includes 3 items. The second subscale evaluates “stress from caring for patients” with 8 items. The third subscale measures “stress from assignments and workload” with 5 items. The fourth subscale focuses on “stress from interactions with teachers and nursing staff” with 6 items. The fifth subscale gauges “stress from the clinical environment” with 3 items. The sixth subscale addresses “stress from peers and daily life” with 4 items. El-Ashry et al. (2022) reported an excellent internal consistency reliability of 0.83 [ 21 ]. Two bilingual translators translated the English version of the scale into Arabic and then back-translated it into English by two other independent translators to verify its accuracy. The suitability of the translated version was confirmed through a confirmatory factor analysis (CFA), which yielded goodness-of-fit indices such as a comparative fit index (CFI) of 0.712, a Tucker-Lewis index (TLI) of 0.812, and a root mean square error of approximation (RMSEA) of 0.100.

Tool three: revised study process questionnaire (R-SPQ-2 F)

It was developed by Biggs et al. (2001). It examines deep and surface learning approaches using only 20 questions; each subscale contains 10 questions [ 22 ]. On a 5-point Likert scale ranging from 0 (never or only rarely true of me) to 4 (always or almost always accurate of me). The total score ranged from 0 to 80, with a higher score reflecting more deep or surface learning approaches. The cut-off points of levels of revised study process questionnaire according to score percentage were low < 33%, moderate 33–66%, and high more than 66%. Biggs et al. (2001) found that Cronbach alpha value was 0.73 for deep learning approach and 0.64 for the surface learning approach, which was considered acceptable. Two translators fluent in English and Arabic initially translated a scale from English to Arabic. To ensure the accuracy of the translation, they translated it back into English. The translated version’s appropriateness was evaluated using a confirmatory factor analysis (CFA). The CFA produced several goodness-of-fit indices, including a Comparative Fit Index (CFI) of 0.790, a Tucker-Lewis Index (TLI) of 0.912, and a Root Mean Square Error of Approximation (RMSEA) of 0.100. Comparative Fit Index (CFI) of 0.790, a Tucker-Lewis Index (TLI) of 0.912, and a Root Mean Square Error of Approximation (RMSEA) of 0.100.

Ethical considerations

The Alexandria University College of Nursing’s Research Ethics Committee provided ethical permission before the study’s implementation. Furthermore, pertinent authorities acquired ethical approval at participating nursing institutions. The vice deans of the participating institutions provided written informed consent attesting to institutional support and authority. By giving written informed consent, participants confirmed they were taking part voluntarily. Strict protocols were followed to protect participants’ privacy during the whole investigation. The obtained personal data was kept private and available only to the study team. Ensuring participants’ privacy and anonymity was of utmost importance.

Tools validity

The researchers created tool one after reviewing pertinent literature. Two bilingual translators independently translated the English version into Arabic to evaluate the applicability of the academic stress and learning approach tools for Arabic-speaking populations. To assure accuracy, two additional impartial translators back-translated the translation into English. They were also assessed by a five-person jury of professionals from the education and psychiatric nursing departments. The scales were found to have sufficiently evaluated the intended structures by the jury.

Pilot study

A preliminary investigation involved 100 nursing student applicants, distinct from the final sample, to gauge the efficacy, clarity, and potential obstacles in utilizing the research instruments. The pilot findings indicated that the instruments were accurate, comprehensible, and suitable for the target demographic. Additionally, Cronbach’s Alpha was utilized to further assess the instruments’ reliability, demonstrating internal solid consistency for both the learning approaches and academic stress tools, with values of 0.91 and 0.85, respectively.

Data collection

The researchers convened with each qualified student in a relaxed, unoccupied classroom in their respective college settings. Following a briefing on the study’s objectives, the students filled out the datasheet. The interviews typically lasted 15 to 20 min.

Data analysis

The data collected were analyzed using IBM SPSS software version 26.0. Following data entry, a thorough examination and verification were undertaken to ensure accuracy. The normality of quantitative data distributions was assessed using Kolmogorov-Smirnov tests. Cronbach’s Alpha was employed to evaluate the reliability and internal consistency of the study instruments. Descriptive statistics, including means (M), standard deviations (SD), and frequencies/percentages, were computed to summarize academic stress and learning approaches for categorical data. Student’s t-tests compared scores between two groups for normally distributed variables, while One-way ANOVA compared scores across more than two categories of a categorical variable. Pearson’s correlation coefficient determined the strength and direction of associations between customarily distributed quantitative variables. Hierarchical regression analysis identified the primary independent factors influencing learning approaches. Statistical significance was determined at the 5% (p < 0.05).

Table  1 presents socio-demographic data for a group of 1010 nursing students. The age distribution shows that 38.8% of the students were between 18 and 21 years old, 32.9% were between 21 and 24 years old, and 28.3% were between 24 and 28 years old, with an average age of approximately 22.79. Regarding gender, most of the students were female (77%), while 23% were male. The students were distributed across different educational years, a majority of 34.4% in the second year, followed by 29.4% in the fourth year. The students’ hours spent studying were found to be approximately two-thirds (67%) of the students who studied between 3 and 6 h. Similarly, sleep patterns differ among the students; more than three-quarters (77.3%) of students sleep between 5- to more than 7 h, and only 2.4% sleep less than 2 h per night. Finally, the student’s Grade Point Average (GPA) from the previous semester was also provided. 21% of the students had a GPA between 2 and 2.5, 40.9% had a GPA between 2.5 and 3, and 38.1% had a GPA between 3 and 3.5.

Figure  1 provides the learning approach level among nursing students. In terms of learning approach, most students (55.0%) exhibited a moderate level of deep learning approach, followed by 25.9% with a high level and 19.1% with a low level. The surface learning approach was more prevalent, with 47.8% of students showing a moderate level, 41.7% showing a low level, and only 10.5% exhibiting a high level.

Nursing students? levels of learning approach (N=1010)

Figure  2 provides the types of academic stress levels among nursing students. Among nursing students, various stressors significantly impact their academic experiences. Foremost among these stressors are the pressure and demands associated with academic assignments and workload, with 30.8% of students attributing their high stress levels to these factors. Challenges within the clinical environment are closely behind, contributing significantly to high stress levels among 25.7% of nursing students. Interactions with peers and daily life stressors also weigh heavily on students, ranking third among sources of high stress, with 21.5% of students citing this as a significant factor. Similarly, interaction with teachers and nursing staff closely follow, contributing to high-stress levels for 20.3% of nursing students. While still significant, stress from taking care of patients ranks slightly lower, with 16.7% of students reporting it as a significant factor contributing to their academic stress. At the lowest end of the ranking, but still notable, is stress from a perceived lack of professional knowledge and skills, with 15.9% of students experiencing high stress in this area.

Nursing students? levels of academic stress subtypes (N=1010)

Figure  3 provides the total levels of academic stress among nursing students. The majority of students experienced moderate academic stress (56.3%), followed by those experiencing low academic stress (29.9%), and a minority experienced high academic stress (13.8%).

Nursing students? levels of total academic stress (N=1010)

Table  2 displays the correlation between academic stress subscales and deep and surface learning approaches among 1010 nursing students. All stress subscales exhibited a negative correlation regarding the deep learning approach, indicating that the inclination toward deep learning decreases with increasing stress levels. The most significant negative correlation was observed with stress stemming from the lack of professional knowledge and skills (r=-0.392, p < 0.001), followed by stress from the clinical environment (r=-0.109, p = 0.001), stress from assignments and workload (r=-0.103, p = 0.001), stress from peers and daily life (r=-0.095, p = 0.002), and stress from patient care responsibilities (r=-0.093, p = 0.003). The weakest negative correlation was found with stress from interactions with teachers and nursing staff (r=-0.083, p = 0.009). Conversely, concerning the surface learning approach, all stress subscales displayed a positive correlation, indicating that heightened stress levels corresponded with an increased tendency toward superficial learning. The most substantial positive correlation was observed with stress related to the lack of professional knowledge and skills (r = 0.365, p < 0.001), followed by stress from patient care responsibilities (r = 0.334, p < 0.001), overall stress (r = 0.355, p < 0.001), stress from interactions with teachers and nursing staff (r = 0.262, p < 0.001), stress from assignments and workload (r = 0.262, p < 0.001), and stress from the clinical environment (r = 0.254, p < 0.001). The weakest positive correlation was noted with stress stemming from peers and daily life (r = 0.186, p < 0.001).

Table  3 outlines the association between the socio-demographic characteristics of nursing students and their deep and surface learning approaches. Concerning age, statistically significant differences were observed in deep and surface learning approaches (F = 3.661, p = 0.003 and F = 7.983, p < 0.001, respectively). Gender also demonstrated significant differences in deep and surface learning approaches (t = 3.290, p = 0.001 and t = 8.638, p < 0.001, respectively). Female students exhibited higher scores in the deep learning approach (31.59 ± 8.28) compared to male students (29.59 ± 7.73), while male students had higher scores in the surface learning approach (29.97 ± 7.36) compared to female students (24.90 ± 7.97). Educational level exhibited statistically significant differences in deep and surface learning approaches (F = 5.599, p = 0.001 and F = 17.284, p < 0.001, respectively). Both deep and surface learning approach scores increased with higher educational levels. The duration of study hours demonstrated significant differences only in the surface learning approach (F = 3.550, p = 0.014), with scores increasing as study hours increased. However, no significant difference was observed in the deep learning approach (F = 0.861, p = 0.461). Hours of sleep per night and GPA from the previous semester did not exhibit statistically significant differences in deep or surface learning approaches.

Table  4 presents a multivariate linear regression analysis examining the factors influencing the learning approach among 1110 nursing students. The deep learning approach was positively influenced by age, gender (being female), educational year level, and stress from teachers and nursing staff, as indicated by their positive coefficients and significant p-values (p < 0.05). However, it was negatively influenced by stress from a lack of professional knowledge and skills. The other factors do not significantly influence the deep learning approach. On the other hand, the surface learning approach was positively influenced by gender (being female), educational year level, stress from lack of professional knowledge and skills, stress from assignments and workload, and stress from taking care of patients, as indicated by their positive coefficients and significant p-values (p < 0.05). However, it was negatively influenced by gender (being male). The other factors do not significantly influence the surface learning approach. The adjusted R-squared values indicated that the variables in the model explain 17.8% of the variance in the deep learning approach and 25.5% in the surface learning approach. Both models were statistically significant (p < 0.001).

Nursing students’ academic stress and learning approaches are essential to planning for effective and efficient learning. Nursing education also aims to develop knowledgeable and competent students with problem-solving and critical-thinking skills.

The study’s findings highlight the significant presence of stress among nursing students, with a majority experiencing moderate to severe levels of academic stress. This aligns with previous research indicating that academic stress is prevalent among nursing students. For instance, Zheng et al. (2022) observed moderated stress levels in nursing students during clinical placements [ 23 ], while El-Ashry et al. (2022) found that nearly all first-year nursing students in Egypt experienced severe academic stress [ 21 ]. Conversely, Ali and El-Sherbini (2018) reported that over three-quarters of nursing students faced high academic stress. The complexity of the nursing program likely contributes to these stress levels [ 24 ].

The current study revealed that nursing students identified the highest sources of academic stress as workload from assignments and the stress of caring for patients. This aligns with Banu et al.‘s (2015) findings, where academic demands, assignments, examinations, high workload, and combining clinical work with patient interaction were cited as everyday stressors [ 25 ]. Additionally, Anaman-Torgbor et al. (2021) identified lectures, assignments, and examinations as predictors of academic stress through logistic regression analysis. These stressors may stem from nursing programs emphasizing the development of highly qualified graduates who acquire knowledge, values, and skills through classroom and clinical experiences [ 26 ].

The results regarding learning approaches indicate that most nursing students predominantly employed the deep learning approach. Despite acknowledging a surface learning approach among the participants in the present study, the prevalence of deep learning was higher. This inclination toward the deep learning approach is anticipated in nursing students due to their engagement with advanced courses, requiring retention, integration, and transfer of information at elevated levels. The deep learning approach correlates with a gratifying learning experience and contributes to higher academic achievements [ 3 ]. Moreover, the nursing program’s emphasis on active learning strategies fosters critical thinking, problem-solving, and decision-making skills. These findings align with Mahmoud et al.‘s (2019) study, reporting a significant presence (83.31%) of the deep learning approach among undergraduate nursing students at King Khalid University’s Faculty of Nursing [ 27 ]. Additionally, Mohamed &Morsi (2019) found that most nursing students at Benha University’s Faculty of Nursing embraced the deep learning approach (65.4%) compared to the surface learning approach [ 28 ].

The study observed a negative correlation between the deep learning approach and the overall mean stress score, contrasting with a positive correlation between surface learning approaches and overall stress levels. Elevated academic stress levels may diminish motivation and engagement in the learning process, potentially leading students to feel overwhelmed, disinterested, or burned out, prompting a shift toward a surface learning approach. This finding resonates with previous research indicating that nursing students who actively seek positive academic support strategies during academic stress have better prospects for success than those who do not [ 29 ]. Nebhinani et al. (2020) identified interface concerns and academic workload as significant stress-related factors. Notably, only an interest in nursing demonstrated a significant association with stress levels, with participants interested in nursing primarily employing adaptive coping strategies compared to non-interested students.

The current research reveals a statistically significant inverse relationship between different dimensions of academic stress and adopting the deep learning approach. The most substantial negative correlation was observed with stress arising from a lack of professional knowledge and skills, succeeded by stress associated with the clinical environment, assignments, and workload. Nursing students encounter diverse stressors, including delivering patient care, handling assignments and workloads, navigating challenging interactions with staff and faculty, perceived inadequacies in clinical proficiency, and facing examinations [ 30 ].

In the current study, the multivariate linear regression analysis reveals that various factors positively influence the deep learning approach, including age, female gender, educational year level, and stress from teachers and nursing staff. In contrast, stress from a lack of professional knowledge and skills exert a negative influence. Conversely, the surface learning approach is positively influenced by female gender, educational year level, stress from lack of professional knowledge and skills, stress from assignments and workload, and stress from taking care of patients, but negatively affected by male gender. The models explain 17.8% and 25.5% of the variance in the deep and surface learning approaches, respectively, and both are statistically significant. These findings underscore the intricate interplay of demographic and stress-related factors in shaping nursing students’ learning approaches. High workloads and patient care responsibilities may compel students to prioritize completing tasks over deep comprehension. This pressure could lead to a surface learning approach as students focus on meeting immediate demands rather than engaging deeply with course material. This observation aligns with the findings of Alsayed et al. (2021), who identified age, gender, and study year as significant factors influencing students’ learning approaches.

Deep learners often demonstrate better self-regulation skills, such as effective time management, goal setting, and seeking support when needed. These skills can help manage academic stress and maintain a balanced learning approach. These are supported by studies that studied the effect of coping strategies on stress levels [ 6 , 31 , 32 ]. On the contrary, Pacheco-Castillo et al. study (2021) found a strong significant relationship between academic stressors and students’ level of performance. That study also proved that the more academic stress a student faces, the lower their academic achievement.

Strengths and limitations of the study

This study has lots of advantages. It provides insightful information about the educational experiences of Egyptian nursing students, a demographic that has yet to receive much research. The study’s limited generalizability to other people or nations stems from its concentration on this particular group. This might be addressed in future studies by using a more varied sample. Another drawback is the dependence on self-reported metrics, which may contain biases and mistakes. Although the cross-sectional design offers a moment-in-time view of the problem, it cannot determine causation or evaluate changes over time. To address this, longitudinal research may be carried out.

Notwithstanding these drawbacks, the study substantially contributes to the expanding knowledge of academic stress and nursing students’ learning styles. Additional research is needed to determine teaching strategies that improve deep-learning approaches among nursing students. A qualitative study is required to analyze learning approaches and factors that may influence nursing students’ selection of learning approaches.

According to the present study’s findings, nursing students encounter considerable academic stress, primarily stemming from heavy assignments and workload, as well as interactions with teachers and nursing staff. Additionally, it was observed that students who experience lower levels of academic stress typically adopt a deep learning approach, whereas those facing higher stress levels tend to resort to a surface learning approach. Demographic factors such as age, gender, and educational level influence nursing students’ choice of learning approach. Specifically, female students are more inclined towards deep learning, whereas male students prefer surface learning. Moreover, deep and surface learning approach scores show an upward trend with increasing educational levels and study hours. Academic stress emerges as a significant determinant shaping the adoption of learning approaches among nursing students.

Implications in nursing practice

Nursing programs should consider integrating stress management techniques into their curriculum. Providing students with resources and skills to cope with academic stress can improve their well-being and academic performance. Educators can incorporate teaching strategies that promote deep learning approaches, such as problem-based learning, critical thinking exercises, and active learning methods. These approaches help students engage more deeply with course material and reduce reliance on surface learning techniques. Recognizing the gender differences in learning approaches, nursing programs can offer gender-specific support services and resources. For example, providing targeted workshops or counseling services that address male and female nursing students’ unique stressors and learning needs. Implementing mentorship programs and peer support groups can create a supportive environment where students can share experiences, seek advice, and receive encouragement from their peers and faculty members. Encouraging students to reflect on their learning processes and identify effective study strategies can help them develop metacognitive skills and become more self-directed learners. Faculty members can facilitate this process by incorporating reflective exercises into the curriculum. Nursing faculty and staff should receive training on recognizing signs of academic stress among students and providing appropriate support and resources. Additionally, professional development opportunities can help educators stay updated on evidence-based teaching strategies and practical interventions for addressing student stress.

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to restrictions imposed by the institutional review board to protect participant confidentiality, but are available from the corresponding author on reasonable request.

Liu J, Yang Y, Chen J, Zhang Y, Zeng Y, Li J. Stress and coping styles among nursing students during the initial period of the clinical practicum: A cross-section study. Int J Nurs Sci. 2022a;9(2). https://doi.org/10.1016/j.ijnss.2022.02.004 .

Saifan A, Devadas B, Daradkeh F, Abdel-Fattah H, Aljabery M, Michael LM. Solutions to bridge the theory-practice gap in nursing education in the UAE: a qualitative study. BMC Med Educ. 2021;21(1). https://doi.org/10.1186/s12909-021-02919-x .

Alsayed S, Alshammari F, Pasay-an E, Dator WL. Investigating the learning approaches of students in nursing education. J Taibah Univ Med Sci. 2021;16(1). https://doi.org/10.1016/j.jtumed.2020.10.008 .

Salah Dogham R, Elcokany NM, Saber Ghaly A, Dawood TMA, Aldakheel FM, Llaguno MBB, Mohsen DM. Self-directed learning readiness and online learning self-efficacy among undergraduate nursing students. Int J Afr Nurs Sci. 2022;17. https://doi.org/10.1016/j.ijans.2022.100490 .

Zhao Y, Kuan HK, Chung JOK, Chan CKY, Li WHC. Students’ approaches to learning in a clinical practicum: a psychometric evaluation based on item response theory. Nurse Educ Today. 2018;66. https://doi.org/10.1016/j.nedt.2018.04.015 .

Huang HM, Fang YW. Stress and coping strategies of online nursing practicum courses for Taiwanese nursing students during the COVID-19 pandemic: a qualitative study. Healthcare. 2023;11(14). https://doi.org/10.3390/healthcare11142053 .

Nebhinani M, Kumar A, Parihar A, Rani R. Stress and coping strategies among undergraduate nursing students: a descriptive assessment from Western Rajasthan. Indian J Community Med. 2020;45(2). https://doi.org/10.4103/ijcm.IJCM_231_19 .

Olvera Alvarez HA, Provencio-Vasquez E, Slavich GM, Laurent JGC, Browning M, McKee-Lopez G, Robbins L, Spengler JD. Stress and health in nursing students: the Nurse Engagement and Wellness Study. Nurs Res. 2019;68(6). https://doi.org/10.1097/NNR.0000000000000383 .

Del Giudice M, Buck CL, Chaby LE, Gormally BM, Taff CC, Thawley CJ, Vitousek MN, Wada H. What is stress? A systems perspective. Integr Comp Biol. 2018;58(6):1019–32. https://doi.org/10.1093/icb/icy114 .

Article   PubMed   Google Scholar  

Bhui K, Dinos S, Galant-Miecznikowska M, de Jongh B, Stansfeld S. Perceptions of work stress causes and effective interventions in employees working in public, private and non-governmental organisations: a qualitative study. BJPsych Bull. 2016;40(6). https://doi.org/10.1192/pb.bp.115.050823 .

Lavoie-Tremblay M, Sanzone L, Aubé T, Paquet M. Sources of stress and coping strategies among undergraduate nursing students across all years. Can J Nurs Res. 2021. https://doi.org/10.1177/08445621211028076 .

Article   PubMed   PubMed Central   Google Scholar  

Ahmed WAM, Abdulla YHA, Alkhadher MA, Alshameri FA. Perceived stress and coping strategies among nursing students during the COVID-19 pandemic: a systematic review. Saudi J Health Syst Res. 2022;2(3). https://doi.org/10.1159/000526061 .

Pacheco-Castillo J, Casuso-Holgado MJ, Labajos-Manzanares MT, Moreno-Morales N. Academic stress among nursing students in a Private University at Puerto Rico, and its Association with their academic performance. Open J Nurs. 2021;11(09). https://doi.org/10.4236/ojn.2021.119063 .

Tran TTT, Nguyen NB, Luong MA, Bui THA, Phan TD, Tran VO, Ngo TH, Minas H, Nguyen TQ. Stress, anxiety and depression in clinical nurses in Vietnam: a cross-sectional survey and cluster analysis. Int J Ment Health Syst. 2019;13(1). https://doi.org/10.1186/s13033-018-0257-4 .

Magnavita N, Chiorri C. Academic stress and active learning of nursing students: a cross-sectional study. Nurse Educ Today. 2018;68. https://doi.org/10.1016/j.nedt.2018.06.003 .

Folkvord SE, Risa CF. Factors that enhance midwifery students’ learning and development of self-efficacy in clinical placement: a systematic qualitative review. Nurse Educ Pract. 2023;66. https://doi.org/10.1016/j.nepr.2022.103510 .

Myers SB, Sweeney AC, Popick V, Wesley K, Bordfeld A, Fingerhut R. Self-care practices and perceived stress levels among psychology graduate students. Train Educ Prof Psychol. 2012;6(1). https://doi.org/10.1037/a0026534 .

Zeb H, Arif I, Younas A. Nurse educators’ experiences of fostering undergraduate students’ ability to manage stress and demanding situations: a phenomenological inquiry. Nurse Educ Pract. 2022;65. https://doi.org/10.1016/j.nepr.2022.103501 .

Centers for Disease Control and Prevention. User Guide| Support| Epi Info™ [Internet]. Atlanta: CDC; [cited 2024 Jan 31]. Available from: CDC website.

Sheu S, Lin HS, Hwang SL, Yu PJ, Hu WY, Lou MF. The development and testing of a perceived stress scale for nursing students in clinical practice. J Nurs Res. 1997;5:41–52. Available from: http://ntur.lib.ntu.edu.tw/handle/246246/165917 .

El-Ashry AM, Harby SS, Ali AAG. Clinical stressors as perceived by first-year nursing students of their experience at Alexandria main university hospital during the COVID-19 pandemic. Arch Psychiatr Nurs. 2022;41:214–20. https://doi.org/10.1016/j.apnu.2022.08.007 .

Biggs J, Kember D, Leung DYP. The revised two-factor study process questionnaire: R-SPQ-2F. Br J Educ Psychol. 2001;71(1):133–49. https://doi.org/10.1348/000709901158433 .

Article   CAS   PubMed   Google Scholar  

Zheng YX, Jiao JR, Hao WN. Stress levels of nursing students: a systematic review and meta-analysis. Med (United States). 2022;101(36). https://doi.org/10.1097/MD.0000000000030547 .

Ali AM, El-Sherbini HH. Academic stress and its contributing factors among faculty nursing students in Alexandria. Alexandria Scientific Nursing Journal. 2018; 20(1):163–181. Available from: https://asalexu.journals.ekb.eg/article_207756_b62caf4d7e1e7a3b292bbb3c6632a0ab.pdf .

Banu P, Deb S, Vardhan V, Rao T. Perceived academic stress of university students across gender, academic streams, semesters, and academic performance. Indian J Health Wellbeing. 2015;6(3):231–235. Available from: http://www.iahrw.com/index.php/home/journal_detail/19#list .

Anaman-Torgbor JA, Tarkang E, Adedia D, Attah OM, Evans A, Sabina N. Academic-related stress among Ghanaian nursing students. Florence Nightingale J Nurs. 2021;29(3):263. https://doi.org/10.5152/FNJN.2021.21030 .

Mahmoud HG, Ahmed KE, Ibrahim EA. Learning Styles and Learning Approaches of Bachelor Nursing Students and its Relation to Their Achievement. Int J Nurs Didact. 2019;9(03):11–20. Available from: http://www.nursingdidactics.com/index.php/ijnd/article/view/2465 .

Mohamed NAAA, Morsi MES, Learning Styles L, Approaches. Academic achievement factors, and self efficacy among nursing students. Int J Novel Res Healthc Nurs. 2019;6(1):818–30. Available from: www.noveltyjournals.com.

Google Scholar  

Onieva-Zafra MD, Fernández-Muñoz JJ, Fernández-Martínez E, García-Sánchez FJ, Abreu-Sánchez A, Parra-Fernández ML. Anxiety, perceived stress and coping strategies in nursing students: a cross-sectional, correlational, descriptive study. BMC Med Educ. 2020;20:1–9. https://doi.org/10.1186/s12909-020-02294-z .

Article   Google Scholar  

Aljohani W, Banakhar M, Sharif L, Alsaggaf F, Felemban O, Wright R. Sources of stress among Saudi Arabian nursing students: a cross-sectional study. Int J Environ Res Public Health. 2021;18(22). https://doi.org/10.3390/ijerph182211958 .

Liu Y, Wang L, Shao H, Han P, Jiang J, Duan X. Nursing students’ experience during their practicum in an intensive care unit: a qualitative meta-synthesis. Front Public Health. 2022;10. https://doi.org/10.3389/fpubh.2022.974244 .

Majrashi A, Khalil A, Nagshabandi E, Al MA. Stressors and coping strategies among nursing students during the COVID-19 pandemic: scoping review. Nurs Rep. 2021;11(2):444–59. https://doi.org/10.3390/nursrep11020042 .

Download references

Acknowledgements

Our sincere thanks go to all the nursing students in the study. We also want to thank Dr/ Rasha Badry for their statistical analysis help and contribution to this study.

The research was not funded by public, commercial, or non-profit organizations.

Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Author information

Authors and affiliations.

Nursing Education, Faculty of Nursing, Alexandria University, Alexandria, Egypt

Rawhia Salah Dogham & Heba Fakieh Mansy Ali

Critical Care & Emergency Nursing, Faculty of Nursing, Alexandria University, Alexandria, Egypt

Nermine M. Elcokany

Obstetrics and Gynecology Nursing, Faculty of Nursing, Alexandria University, Alexandria, Egypt

Asmaa Saber Ghaly

Faculty of Nursing, Beni-Suef University, Beni-Suef, Egypt

Mohamed Mahmoud Seweid

Psychiatric and Mental Health Nursing, Faculty of Nursing, Alexandria University, Alexandria, Egypt

Ayman Mohamed El-Ashry

You can also search for this author in PubMed   Google Scholar

Contributions

Ayman M. El-Ashry & Rawhia S. Dogham: conceptualization, preparation, and data collection; methodology; investigation; formal analysis; data analysis; writing-original draft; writing-manuscript; and editing. Heba F. Mansy Ali & Asmaa S. Ghaly: conceptualization, preparation, methodology, investigation, writing-original draft, writing-review, and editing. Nermine M. Elcokany & Mohamed M. Seweid: Methodology, investigation, formal analysis, data collection, writing-manuscript & editing. All authors reviewed the manuscript and accept for publication.

Corresponding author

Correspondence to Ayman Mohamed El-Ashry .

Ethics declarations

Ethics approval and consent to participate.

The research adhered to the guidelines and regulations outlined in the Declaration of Helsinki (DoH-Oct2008). The Faculty of Nursing’s Research Ethical Committee (REC) at Alexandria University approved data collection in this study (IRB00013620/95/9/2022). Participants were required to sign an informed written consent form, which included an explanation of the research and an assessment of their understanding.

Consent for publication

Not applicable.

Competing interests

The authors declare that there is no conflict of interest.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Dogham, R.S., Ali, H.F.M., Ghaly, A.S. et al. Deciphering the influence: academic stress and its role in shaping learning approaches among nursing students: a cross-sectional study. BMC Nurs 23 , 249 (2024). https://doi.org/10.1186/s12912-024-01885-1

Download citation

Received : 31 January 2024

Accepted : 21 March 2024

Published : 17 April 2024

DOI : https://doi.org/10.1186/s12912-024-01885-1

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Academic stress
• Learning approaches
• Nursing students

BMC Nursing

ISSN: 1472-6955

• General enquiries: [email protected]

• Our Mission

PBL for Pre-K Through Second Grade

Very young students can benefit from project-based learning, as these detailed steps for a project conducted by preschool students demonstrate.

Observation, exploration, and discovery are three main skills that young children (kindergarten to second grade) generally develop when they interact with their surroundings. Some children prefer to take time to observe the environment before moving around to explore, while others choose to immediately start actively discovering the wonders within the environment. Nurturing an environment that ignites curiosity and facilitates exploration, therefore, is paramount.

Early-years educators who work with children 2 to 7 years old play a pivotal role in laying the foundation for lifelong learning by creating spaces where children can freely explore their diverse interests and learn how to expand explorations and inquiries into meaningful in-depth projects. 

In this post, I share a project I developed for a pre-K class with 3-year-olds that offers clear examples of each step and process feature. The project began when a boy became fascinated by the sound that came out of a bottle when he blew into it. He shared his findings with his peers, and the Sounds Exploration project began.  

Creating an Environment for Exploration

The environments where young children interact become learning spaces and serve as educators, generating dialogue between children and/or between each child and the environment, developing processes of inquiry, reflection, observation, and active listening. What should these spaces be like? 

Versatility: Design flexible learning environments that allow spontaneous exploration and discovery in different areas and disciplines. Incorporate adaptable learning materials to accommodate diverse interests and activities. Equip your classroom with a rich variety of resources, including books, art supplies, natural materials, and sensory experiences. Allow children to access natural open spaces that offer enough room for free exploration.  

In the example of the Sounds Exploration project, the teacher offered different materials and contexts for the learners to make and identify sounds, both outside and indoors. In the open air, the teacher helped learners focus on sounds by inviting them to close their eyes and name the sounds they could hear. This takes some time and guidance to help learners to listen beyond the sounds in the foreground and pay attention to those in the background. The learners were then made aware of the sounds they made by walking on different surfaces—like pebbles, grass, and mud—and the sounds they made with sticks or by hitting the water. Outdoors contexts are full of discovery possibilities. 

Indoors, the children used materials such as plastic tops, foil, plastic, cellophane, construction paper, and cardboard to make sounds. Wood blocks, musical instruments, and toys (cars, trucks, dolls, balls, construction blocks) are also an option, as are the different surfaces (floor, carpets, tables) in the classroom. 

In both environments, the children had guidance to help them discover more about the sounds they produced. This connects with the next point.

Curiosity: Encourage a culture of inquiry by posing open-ended questions, stimulating wonder, and inviting children to explore topics of interest. Offer provocations and invitations to learning that spark curiosity and prompt further investigation. 

Collaboration: Facilitate opportunities for children to explore together, interact, learn from one another, and question their findings. 

Interacting with children during exploration periods

The most successful interactions a teacher can carry out in the period of inquiry are those that don’t have a single answer but allow for different responses. The children’s answers will likely be the result of the connections they make with themselves, their previous knowledge, their interaction with their peers, and the context in which they’re interacting.

As a consequence, the teacher has an opportunity to develop and model an attitude of listening and of inquiry into the children’s responses and the construction of their learning. 

For example, related to the Sounds project, the teacher might ask the following questions: 

• “How do you do it? Can you teach me?”
• “This sound... what does it remind you of?” 
• ”What other sounds can we make?”
•  ”What causes sound to be produced?”
•  ”What can we use this sound for?”

Engage in active observation: Observe children closely as they play: as they interact with each other, the decisions they make, and how they choose to communicate their feelings, emotions, thoughts. Pay attention to their interests, preferences, and inquiries.

Listen actively: Listen to the children’s conversations when you ask open-ended questions to stimulate their thinking and foster reflection and critical thinking. Encourage them to communicate their ideas and their thoughts, share observations, and voice their desire to know. 

Let the children express freely: Let them show you their willingness to deepen their knowledge. Follow their interests and curiosity, allowing them to guide the direction of their exploration. Facilitate support and resources based on their inquiries, empowering them to build knowledge and take ownership of their learning journey.

Provide research tools: Offer the children access to age-appropriate tools and materials, including books, digital resources, and hands-on experiences. Support them in navigating these resources independently, fostering self-directed learning skills.

Facilitate tools to document their findings: Provide materials and resources for learners to document their discoveries in various ways: different art forms, notes, oral dialogues, audio/video recordings. 

Transferring exploration into research projects

Children’s active exploration, properly documented, will generate a lot of information and, in turn, will create the possibility of continuing work on a specific project.

In the example of the Sounds Exploration project, the learners were invited to use the sounds they had collected, identified, and documented to make a Sound Story from a well-known story they usually read in class and enjoyed. The guiding question was this: How can the learners in this class turn [the name of the story] into a sound story?

Assist project planning: Guide children in planning and organizing their research project, and deconstruct the process into manageable steps. Help them create research questions, collect information, and develop a short-term plan of action. 

Analyze the data collected: Facilitate understanding of the findings and guide the children to become aware of which subject area they’re willing to learn more about. 

Ignite Intrinsic motivation: Provide steps for the learners to become aware of  what they already know about the specific topic in that subject area and what more they want to know, and guide them in finding where they can collect the information they’re looking for. 

Foster reflection: Promote reflection throughout the research process. Provide opportunities for children to share their findings with peers and reflect on their learning experiences and strategies.

Research projects enable teachers to empower children to make choices and decisions about their learning journey when they have a range of options and opportunities to explore their interests authentically. In addition, research projects foster collaboration and peer learning by encouraging children to work together and share what they’ve learned. 

It’s important to recognize and celebrate children’s achievements and contributions throughout the research process. Create opportunities for them to showcase their work, share their findings with others, and receive feedback and praise .

In essence, by creating an environment that nurtures exploration, supporting children during their inquiries, and empowering them to take on leadership roles in their learning, early years educators can lay the groundwork for a lifetime of curiosity, discovery, and success.

• SUGGESTED TOPICS
• The Magazine
• Newsletters
• Managing Yourself
• Managing Teams
• Work-life Balance
• The Big Idea
• Data & Visuals
• Reading Lists
• Case Selections
• HBR Learning
• Topic Feeds
• Account Settings
• Email Preferences

6 Common Leadership Styles — and How to Decide Which to Use When

• Rebecca Knight

Being a great leader means recognizing that different circumstances call for different approaches.

Research suggests that the most effective leaders adapt their style to different circumstances — be it a change in setting, a shift in organizational dynamics, or a turn in the business cycle. But what if you feel like you’re not equipped to take on a new and different leadership style — let alone more than one? In this article, the author outlines the six leadership styles Daniel Goleman first introduced in his 2000 HBR article, “Leadership That Gets Results,” and explains when to use each one. The good news is that personality is not destiny. Even if you’re naturally introverted or you tend to be driven by data and analysis rather than emotion, you can still learn how to adapt different leadership styles to organize, motivate, and direct your team.

Much has been written about common leadership styles and how to identify the right style for you, whether it’s transactional or transformational, bureaucratic or laissez-faire. But according to Daniel Goleman, a psychologist best known for his work on emotional intelligence, “Being a great leader means recognizing that different circumstances may call for different approaches.”

• RK Rebecca Knight is a journalist who writes about all things related to the changing nature of careers and the workplace. Her essays and reported stories have been featured in The Boston Globe, Business Insider, The New York Times, BBC, and The Christian Science Monitor. She was shortlisted as a Reuters Institute Fellow at Oxford University in 2023. Earlier in her career, she spent a decade as an editor and reporter at the Financial Times in New York, London, and Boston.

Partner Center

« RETURN TO NEWS

Merck, A&T launch joint Merck Biotechnology Learning Center

04/19/2024 Research and Economic Development

• 4-H and Youth Development News
• Academic Affairs News
• Accounting and Finance News
• Administration and Instructional Services News
• Admissions News
• Agribusiness, Applied Economics and Agriscience Education News
• Agricultural and Natural Resources News
• Alumni News
• Animal Sciences News
• Applied Engineering Technology News
• Athletics News
• Biology News
• Built Environment News
• Business and Finance News
• Business Education News
• Chancellor's Speaker Series
• Chancellors Town Hall Series
• Chemical, Biological, and Bio Engineering News
• Chemistry News
• Civil, Architectural and Environmental Engineering News
• College News
• Community and Rural Development News
• Computational Science and Engineering News
• Computer Science News
• Computer Systems Technology News
• Cooperative Extension News
• Counseling News
• Criminal Justice News
• Deese College News
• Economics News
• Educator Preparation News
• Electrical and Computer Engineering News
• Employees News
• Energy and Environmental Systems News
• English Department News
• Family and Consumer Sciences News
• Graphic Design Technology News
• Hairston College News
• Headlines News
• History & Political Science News
• Honors College News
• Human Resources News
• Industrial and Systems Engineering News
• Information Technology Services News
• Innovation Station News
• Journalism & Mass Communication
• Kinesiology News
• Leadership Studies and Adult Education News
• Liberal Studies News
• Library News
• Magazine News
• Management News
• Marketing News
• Mathematics News
• Mechanical Engineering News
• Media Spotlight News
• Natural Resources and Environmental Design News
• News Categories
• Nursing News
• Psychology News
• Research and Economic Development News
• Social Work News
• Student Affairs News
• Students News
• The Graduate College News
• Transportation & Supply Chain
• University Advancement News
• Visual & Performing Arts News

Collaboration supports student enrichment and workforce development in North Carolina

EAST GREENSBORO, N.C. (April 19, 2024) – Merck (NYSE: MRK), known as MSD outside of the United States and Canada, and North Carolina Agricultural and Technical State University announced today the opening of the Merck Biotechnology Learning Center at Gateway Research Park in Greensboro, North Carolina.

The Merck Biotechnology Learning Center is a 4,025-square-foot facility that includes classroom space, a process laboratory and state-of-the-art biopharmaceutical manufacturing equipment. In the Learning Center, students and Merck trainees will experience hands-on learning and advanced discovery opportunities to enhance academic programming and training for biotechnology careers.

"We are embarking on a significant journey with the launch of the Merck Biotechnology Learning Center and our collaboration with N.C. A&T,” said Sanat Chattopadhyay, executive vice president and president, Merck Manufacturing Division. “The Learning Center is not just a building; it's an incubator for innovation, a path to discovery, and a beacon guiding the next generation of thinkers, problem-solvers and leaders who will drive our industry forward.”

The opening was marked with a joint celebration that included senior leaders from both Merck and A&T, current and former A&T students, and local government officials, including North Carolina Commerce Secretary Machelle Baker Sanders. Attendees participated in tours of the lab and classroom facilities to see firsthand the immersive learning opportunities.

Merck and A&T, America’s premier historically Black doctoral research university, developed this collaboration based on mutual values of innovation, community engagement and a commitment to diversity and inclusion. The joint effort between Merck and A&T supports the increasing need for biotech training and education in North Carolina and highlights the importance of business and historically Black college and university (HBCU) cooperation in growing diverse talent in the biotech sector.

“The Merck Biotechnology Learning Center will provide opportunities for N.C. A&T students to understand what a career in biotech looks like,” said Amanda Taylor, vice president and plant manager at the Merck Manufacturing Division site in Durham, North Carolina. “We have several wonderful N.C. A&T graduates working at our Durham site already, and there is so much growth in manufacturing across North Carolina. Through our collaboration with N.C. A&T, we’re developing new and innovative ways to build a pipeline of talent in the Triad and beyond.”

The opening of the Merck Biotechnology Learning Center is the launch of a long-term collaboration between Merck and A&T. The two organizations will partner on several initiatives to support student enrichment, including curricula development, a speaker series and STEM (Science, Technology, Engineering and Mathematics) community outreach.

“I am thrilled to announce our groundbreaking collaboration with Merck, which heralds a new era of innovation in biotechnology education," said Tonya Smith-Jackson, Ph.D., provost and executive vice chancellor of Academic Affairs. “This partnership signifies a union between academia and industry, and a commitment to excellence, innovation and the advancement of scientific knowledge, as it not only provides our students with unparalleled access to state-of-the-art labs, but also invaluable mentorship from Merck professionals, ensuring they emerge as industry-ready leaders poised to shape the future of biotechnology. The Merck Biotechnology Learning Center will serve as a hub of learning and discovery, and it is also the start of a collaboration where we are going to jointly advance the mission of both Merck and N.C. A&T.”

Merck has been a member of the North Carolina community for more than 40 years. Today, nearly 1,500 Merck colleagues work at North Carolina facilities in Durham and Wilson, including numerous A&T alumni. A&T's College of Engineering is the No. 1 producer of African American graduates in engineering in the United States.

Media Contact Information: [email protected]

Latest News

04/19/2024 in Research and Economic Development

Cheerleaders Earn Historic National Title after “Road 2 Redemption”

04/18/2024 in Athletics , Students

A&T Receives 2024 IIE American Passport Project Grant for Study Abroad

04/12/2024 in Academic Affairs

• Make a Gift
• Directories

Search form

You are here.

• Autumn 2024

ENGL 108 A: Writing Ready: Preparing for College Writing

•   Instagram
•   Newsletter

Multi-factor stock trading strategy based on DQN with multi-BiGRU and multi-head ProbSparse self-attention

• Open access
• Published: 22 April 2024

Cite this article

You have full access to this open access article

• Wenjie Liu 1 , 2 , 3 ,
• Yuchen Gu   ORCID: orcid.org/0000-0002-7267-3541 1 &
• Yebo Ge 1  

Reinforcement learning is widely used in financial markets to assist investors in developing trading strategies. However, most existing models primarily focus on simple volume-price factors, and there is a need for further improvement in the returns of stock trading. To address these challenges, a multi-factor stock trading strategy based on Deep Q-Network (DQN) with Multi-layer Bidirectional Gated Recurrent Unit (Multi-BiGRU) and multi-head ProbSparse self-attention is proposed. Our strategy comprehensively characterizes the determinants of stock prices by considering various factors such as financial quality, valuation, and sentiment factors. We first use Light Gradient Boosting Machine (LightGBM) to classify turning points for stock data. Then, in the reinforcement learning strategy, Multi-BiGRU, which holds the bidirectional learning of historical data, is integrated into DQN, aiming to enhance the model’s ability to understand the dynamics of the stock market. Moreover, the multi-head ProbSparse self-attention mechanism effectively captures interactions between different factors, providing the model with deeper market insights. We validate our strategy’s effectiveness through extensive experimental research on stocks from Chinese and US markets. The results show that our method outperforms both temporal and non-temporal models in terms of stock trading returns. Ablation studies confirm the critical role of LightGBM and multi-head ProbSparse self-attention mechanism. The experiment section also demonstrates the significant advantages of our model through the presentation of box plots and statistical tests. Overall, by fully considering the multi-factor data and the model’s feature extraction capabilities, our work is expected to provide investors with more precise trading decision support.

Graphical abstract

Avoid common mistakes on your manuscript.

1 Introduction

Financial markets are regarded as a significant driving force in the growth and evolution of the global economy. Investing in the stock market is a prominent research field in the financial industry, which involves high returns as well as high risks. The stock market is affected by a variety of factors, such as the corporate finance [ 9 ], exchange rates [ 22 ] and public sentiment [ 3 ]. With the development of technology, quantitative trading plays an important role in stock market investment. The main motivation to promote the development of quantitative trading is to achieve higher and more stable returns in the investment process.

With the continuous development of machine learning, various neural network models have been applied to stock market prediction tasks, including stock price prediction [ 7 , 17 , 25 ], stock trend prediction [ 2 , 19 ], portfolio management [ 1 , 31 ], trading strategies [ 4 , 8 , 26 , 28 , 34 ], etc. The goal of stock price prediction is to determine the future price of stocks, and prediction models typically combine technical analysis and fundamental analysis with external factors such as market sentiment and news events to improve the accuracy of forecasts. Stock trend prediction involves classifying stock trends into three categories: upward, downward, and sideways trends, this typically involves analyzing historical price data to identify trend patterns and the market factors that support the continuation of these trends. Portfolio management involves asset allocation and risk management to optimize the investment portfolio. Quantitative trading strategies use mathematical and statistical methods to identify opportunities in the market and design algorithms to automatically execute trades.

In the field of stock trading strategies, various applications of deep learning have shown significant potential and advantages. Fister et al. [ 8 ] tried to construct a superior strategy for the daily trading on a portfolio of stocks using Long Short-Term Memory (LSTM), which outperforms traditional trading strategies. And then, a hybrid convolutional recurrent neural network was proposed to construct trading strategies to achieve better returns by Yu and Li [ 34 ], where Convolutional Neural Network (CNN) is used to capture local fluctuation features, and LSTM is used to learn the long-term temporal dependencies to improve stock performance prediction. Touzani and Douzi [ 28 ] proposed a trading strategy designed for the Moroccan stock market, the decision-making rules for stock trading are based on predictions given by LSTM and Gate Recurrent Unit (GRU), where LSTM and GRU predict short-term and medium-term closing prices, respectively. Soleymani and Paquet [ 24 ] developed a portfolio management framework based on the DeepBreath deep reinforcement learning framework, which integrates a restricted stacked autoencoder and a CNN module into a unified framework, they also tested the performance of the DeepBreath framework, achieving investment returns that outperform current expert investment strategies while minimizing market risk.

Deep Q network (DQN) is a type of deep reinforcement learning algorithm that has been widely used to solve various challenging tasks. In recent years, with the development of financial technology, the application of DQN in stock trading strategies has also received widespread attention. In 2020, Lei et al. [ 13 ] proposed a Time-driven Feature-aware Jointly Deep Reinforcement Learning model (TFJ-DRL) which is evaluated on real-world financial data with different price trends. In 2021, Ma et al. [ 18 ] proposed a novel model named Parallel Multi-Module deep Reinforcement Learning (PMMRL) algorithm, where Fully Connected (FC) layers and LSTM layers are used to extract and encode features of stocks. In 2023, Wang et al. [ 29 ] used Light Gradient Boosting Machine (LightGBM) to extract and classify turning points in stock prices, and employed DQN and its related models to train and form a set of reinforcement learning trading strategies. In identical year, Liu et al. [ 16 ] proposed a Multi-type data Fusion framework with Deep Reinforcement Learning (MSF-DRL) that integrate stock data, technical indicators, and candlestick charts for algorithmic trading, this approach extracts and fuses temporal features from these data sources to aid in making trading decisions. Although many of the above methods are used to construct trading strategies, the returns achieved are slightly low and the feature extraction ability of the model can be further improved.

Based on the DQN model, we propose a multi-factor stock trading strategy based on DQN with Multi-BiGRU and multi-head ProbSparse self-attention. The main contributions include three aspects:

To characterize stock prices from multiple perspectives, a new multi-factor strategy, including financial quality factors, valuation factors, sentiment factor, etc., is adopted, this approach enhances predictive capabilities and adapts to market changes.

The DQN with Multi-BiGRU is designed, aiming to enhance the model’s ability to understand the dynamics of the stock market, ultimately leading to the generation of more effective and robust trading strategies.

The multi-head ProbSparse self-attention is designed to effectively capture the long-term dependencies between multiple factors and help the model adapt to dynamically changing financial market data.

Compared with state-of-the-art algorithms in recent years, the proposed model based on DQN with Multi-BiGRU and multi-head ProbSparse self-attention achieves higher returns in individual stocks across Chinese and American markets. This significant advantage is also confirmed by ablation experiments, statistical tests, and the boxplots of the experimental results.

The rest of this paper is organized as follows. First, we give a brief related work on DQN and attention mechanisms in stock trading strategies in Section 2 . In Section 3 , we introduce LightGBM, BiGRU, DQN, and the stock market terms used in our research. Then, we illustrate the formulation process of the proposed strategy in Section 4 . Afterwards, in the Section 5 , we make an empirical study on the strategy and discusses its effectiveness. Finally, Section 6 summarizes this paper and discusses the future research direction.

2 Related work

2.1 dqn with non-temporal models.

As a typical reinforcement learning algorithm, DQN [ 21 ] has been concerned by many scholars, which is widely used in games, robot control, autonomous driving and other fields. In recent years, DQN has become possible in the field of stock trading. Li et al. [ 14 ] proposed a novel stock trading strategy consisting of DQN, Double DQN, and Dueling DQN respectively, and found that the DQN model performed best in dealing with stock market strategy decision-making problems. Shi et al. [ 23 ] proposed a stock price trend prediction and stock trading model based on Double DQN, and added a well-designed CNN layer to the strategy network to further extract and leverage potential dependencies in stock data, thereby improving model stability. Based on using DQN, Liu et al. [ 15 ] took into account the sentiment of investors in the forecast of the stock market trend, and dynamically adjusted the length of the input market data according to the market conditions. Takara et al. [ 27 ] proposed a deep reinforcement learning model called Extended Trading DQN (ETDQN) based on DQN, ETDQN prioritizes experiences that include different sub-goals, allowing the model to accumulate maximum profit without the need for complex reward fine-tuning. Wang et al. [ 29 ] used LightGBM to classify turning points in stocks, then used DQN to construct trading strategies, and finally combined the two parts. In these DQN architectures, the Q networks, such as CNNs, automatically extract local features from input data through convolutional layers and pooling layers. However, for stock time series data, their performance in handling long-term dependencies is limited. Therefore, in our selection of Q networks, we have chosen temporal networks. In our comparative experiments, we compared the experimental results with both temporal and non-temporal networks.

2.2 DQN with temporal models

Considering the time series data of investment products such as stocks, many scholars have introduced time series neural networks such as Recurrent Neural Network (RNN) into the DQN model. Wu et al. [ 32 ] used GRU to extract financial information characteristics, and proposed two trading strategies with reinforcement learning methods as Gated Deep Q-Network trading strategy (GDQN) and Gated Deterministic Policy Gradient trading strategy (GDPG), then tested stocks from different countries. Huang et al. [ 11 ] proposed a novel deep reinforcement learning algorithm called dubbed Two-Branch Deep Q-Network (TBDQN), for generating reliable trading signals for crude oil and natural gas futures. In their another work, Huang et al. [ 12 ] proposed a novel algorithmic trading method called CR-DQN, which integrates deep Q-learning with two popular trading rules: Moving Average (MA) and Trading Range Breakout (TRB). The authors have also designed a reward-driven scheme, aiming to capture the intrinsic features of financial data. Lei et al. [ 13 ] proposed TFJ-DRL that combines supervised deep learning and reinforcement learning to enhance feature representation learning and decision-making in financial trading, utilizing attention mechanisms and iterative training to optimize investment returns. Ma et al. [ 18 ] proposed a novel model called PMMRL algorithm, in which one module employs FC layers to learn the current market state from the market data of the traded stocks and the fundamental data of the issuing companies; another module uses LSTM layers to capture the long-term historical trends of the market. Liu et al. [ 16 ] proposed MSF-DRL, in which LSTM, CNN and BiLSTM networks are used to extract the temporal features of stock data and the candlestick chart. In the above models, for diverse and changing stock data, the feature extraction ability of the model can be further improved. Considering that future prices are influenced by past prices and the importance of trend identification in the stock market, we introduced Multi-BiGRU to better learn the forward and backward dependencies in sequential data and capture long-term dependencies in the sequence data when processing stock multi-factor data.

2.3 Attention mechanism

To capture key information and enable adaptive learning, the attention mechanism has been proposed and is increasingly being used in tasks related to the stock market. Han et al. [ 10 ] proposed an Industry Relationship-Driven Hypergraph Attention Network (IRD-HGAT) to predict stock price movement trends. The hypergraph attention mechanism is utilized to dynamically adjust the interactions between stocks, while the features of industry hyperedges are combined to gauge the influence of industry relationships on stock prices. Yang et al. [ 33 ] proposed a stock prediction model MDF-DMC that combines multi-view stock data features with dynamic market-related information, multi-head attention mechanisms and position masking matrices are used to capture the relevance between the stock to be predicted and the market. Cui et al. [ 6 ] proposed a novel hybrid model named McVCsB, which integrates multi-channel VMD-CBAM-BiLSTM, based on an improved decomposition-reconstruction prediction framework for stock index forecasting, the model utilizes the Variational Modal Decomposition algorithm (VMD) and the Convolutional Block Attention Module (CBAM) to achieve noise filtering and deep-level feature extraction. To effectively capture the important features and their nonlinear relationships among stock multi-factor data, we introduced multi-head ProbSparse self-attention. Multi-head ProbSparse self-attention introduces a probabilistic mechanism to sparsify attention weights, reducing the model’s focus on irrelevant information, thereby enhancing efficiency and robustness in processing dynamically changing financial market data.

Structure diagram of GRU

Schematic diagram of DQN

3 Preliminaries

3.1 lightgbm.

Light gradient boosting machine (LightGBM) [ 20 ] is a gradient boosting model based on decision trees, which has the advantages of efficient parallel training, feature parallelism, and fast processing of massive data. LightGBM is used for various machine learning tasks such as sorting, classification, and regression. The optimization of LightGBM is as follows:

Unilateral gradient sampling algorithm is used where only data with high gradients is used when calculating information gain, reducing time overhead.

Mutually exclusive feature bundling is used which can bind multiple mutually exclusive features into one feature, achieving the goal of dimensionality reduction.

Histogram algorithm is used to obtain sub optimal solutions for segmented features and segmentation points, and subtracting the histogram of a child node from the histogram of a parent node, the histogram of another child node can be obtained, reducing the computational complexity by half.

The leaf growth strategy of Leaf-wise with depth limitation is used to split only the points with the maximum information gain, so as to avoid overfitting.

Gated recurrent unit (GRU) [ 5 ] is a variant of the LSTM network. It is a very popular sequential neural network. It can solve the long-term dependency problem in the RNN network. The structure of GRU is is shown in Fig. 1 .

Process of the proposed stock trading strategy based on DQN with Multi-BiGRU and multi-head ProbSparse self-attention

GRU introduces the concepts of reset gate and update gate, the inputs for reset and update (Fig. 2 ) gates are both the input for the current time step \({x_t}\) and the hidden state of the previous time step \({h_{t - 1}}\) . The mathematical expression of update and reset gates are as follows:

where \({z_t}\) and \({r_t}\) are the the update gate and reset gate, \({W_z}\) and \({W_r}\) are the weight matrixs, \({x_t}\) is the input for the current time step, \({h_{t - 1}}\) is the hidden state of the previous time step, and \({\sigma }\) represents the sigmoid function.

After calculating the update gate and reset gate, GRU will calculate \(\widetilde{h}_t\) , and the calculation formula for candidate hidden states is as follows:

Where \({\widetilde{h}_t}\) is the candidate hidden states. The calculation formula for the output \({h_t}\) of GRU at the last time t is as follows:

Reinforcement learning is inspired by behaviorist psychology, focusing on online learning and attempting to keep balance between exploration and utilization. Q-Learning [ 30 ] is a value based algorithm in Reinforcement learning, the simplest implementation of Q-learning is to store reward values in a Q-table, but this approach is limited by the number of states and action spaces. To solve this problem, deep neural network was introduced into Q learning by the DeepMind team [ 21 ], called deep Q network (DQN), in which deep neural network is used as a function approximator to effectively process high-dimensional data and extract key features.

The network serves as an approximation function to calculate the action values of a series of predefined actions for a given state variable s as a cumulative future reward. The approximate function is expressed as Q(s,a). When selecting actions based on the network, the actions with the maximum Q value in the current state s are selected as further actions. The target value function for DQN is as follows:

where \({R_t}\) is the reward obtained after taking action a in the current state, \(\gamma \) is the discount factor, \(s^{\prime }\) is the next state, and \(\theta ^{\prime }\) is the parameter of the target network.

3.4 Glossary

In the subsequent content of the article, particularly in the section describing multi-factor data, some stock terms and proper nouns are involved. For example, NP represents Net profit, and OI represents Operating income. A complete list of stock terms and their corresponding definitions are provided in Appendix A .

4 Multi-factor stock trading strategy based on DQN with Multi-BiGRU and Multi-head ProbSparse Self-attention

4.1 overview.

The overall framework of the stock trading strategy proposed in this paper is shown in Fig. 3 . It can be summarized as follows:

Multiple factors acquisition We calculate ATI &RTI (absolute technical indicators, relative technical indicators), OCHLVC (open, close, high, low, volume and change) and multiple factors, then the comment text of individual stocks is crawled through forums, and their emotional values are obtained through the BiLSTM model.

Data fusion We fuse multiple factors, OCHLVC and sentiment values and use the z-score method to normalize them.

LightGBM model The LightGBM model is used to obtain the turning point buying signal based on the input ATI &RTI.

DQN with Multi-BiGRU and multi-head ProbSparse self-attention After receiving multiple factors as input, the trained reinforcement learning model obtains Q values corresponding to different actions to generate reinforcement learning trading strategies.

Strategy combination If the result of turning point buying signal is true (non-top turning point) and the decision result of RL strategy is ’buy’, the transaction decision should be to buy, while the stock will be sold on the next day by default.

4.2 Multiple factors processing and Sentiment analysis

4.2.1 multiple factors processing.

In this paper, we select multiple factors that affect the stock prices. The types of factors are described in Table 1 :

After obtaining the basic volume-price factors and financial factors, we can calculate more valuable multiple factors based on the formulas. For example, the turnover rate in the past month can be obtained through volume and outstanding shares (OS).

Absolute technical indicators, relative technical indicators and complete multiple factors’ definitions are shown in Appendix B .

Then we fuse multiple factors, OCHLVC and sentiment values and use the z-score method to normalize them. The normalization fomula is as follows.

where z is the value after normalization, x represents the value before normalization, \(\mu \) represents the overall average, and \(\sigma \) represents the overall standard deviation.

Process of the sentiment analysis

Model structure of DQN with Multi-BiGRU

4.2.2 Sentiment analysis

We crawled the comment text data of individual stocks over the past decade from economic forums such as Eastmoney, Snowball Finance. The comment text is passed into the BiLSTM model to analyze sentiment values for one of the input data of the reinforcement learning model. The model structure is shown in Fig. 4 .

Data preprocessing Considering that sentiment during trading hours is mostly (Fig. 5 ) reflected in stock price data, comment text outside trading hours is selected.

CBOW Compared to skipgram, CBOW trains faster and works better for frequently occurring words. We use the CBOW-based training model as the foundation for training word vectors, where \({w_t}\) represents the t-th word in the text, and \({w_t} \in {R^{\mathrm{{ }}d}}\) , then the word vector matrix M can be represented as:

After text word vectorization, the word vector matrix M is input into the BiLSTM.

BiLSTM layer BiLSTM is composed of two LSTMs in different directions, which can better capture bidirectional semantic features. The output of BiLSTM at time t concatenates the outputs of forward and backward LSTM at time t using vectors.

where \({\overrightarrow{{h_t}}}\) and \({\overleftarrow{{h_t}}}\) are the outputs of forward and backward LSTM, W , U , and V represent the weight matrices for forward computation, and W’ , U’ , and V’ represent the weight matrices for backward computation.

Sentiment classification Transfer the output to the fully connected layer for sentiment label prediction, and use the Softmax function to obtain the sentence level sentiment probability distribution. The function is defined as:

where \({y_t}\) is the output value of the t -th node, C is the number of output nodes.

4.3 DQN with Multi-BiGRU

Considering the time-series characteristics of the stock price, the Multi-BiGRU network is integrated into DQN as the Q network, allowing the model to fully consider all information before the current day while observing locally, in order to improve the model’s prediction accuracy.

In a single GRU, due to the inconsistency between the length of the input vector and the length of the hidden layer feature values, w is separated by the input and hidden states in each formula. The mathematical expressions for the GRU unit are as follows:

where \({r_t}\) , \({z_t}\) and \({n_t}\) are reset gates, update gates, and status thresholds in GRU, \({h_t}\) and \({h_{t - 1}}\) are the current hidden state and the previous hidden state.

We combine bidirectional GRU with stacked GRU. In the general case of constructing a network, the hidden state of bidirectional GRU depends on the input value and the previous hidden state; Unlike bidirectional processing, the hidden state of stacked GRUs depends on the previous hidden state and the current hidden state of the previous layer. In Multi-BiGRU, we stack two layers of GRUs in both the forward and backward layers. The mathematical expressions for the first layer BiGRU network structure are as follows:

where \(\overrightarrow{{h_{{1_t}}}} \) and \(\overleftarrow{{h_{{1_t}}}} \) are the forward and backward hidden layer states at time t , \(\overrightarrow{{h_{{1_{t - 1}}}}} \) and \(\overleftarrow{{h_{{1_{t - 1}}}}} \) are previous hidden states in the current layer. ’;’ represents vector concatenation.

In addition to the previous hidden state, the forward and backward networks of the second BiGRU receive the outputs of the first BiGRU’s forward and backward networks respectively, and concatenate the forward and backward vectors to complete the information fusion. Finally, they are fed into the linear layer for output mapping. The mathematical expressions for the second layer BiGRU network structure are as follows:

where \(\overrightarrow{{h_{{2_t}}}} \) and \(\overleftarrow{{h_{{2_t}}}} \) are the hidden states output by the first BiGRU layer, \(\overrightarrow{{h_{{2_{t - 1}}}}} \) and \(\overleftarrow{{h_{{2_{t - 1}}}}} \) are previous hidden states in the current layer.

From the output of the Multi-BiGRU network, we can obtain the estimated output value for the next state. The fitting target of DQN can be seen as a regression problem, mean square error is used as the loss function to minimize the square loss. The loss function is as follows.

where \({R_t}\) is the reward obtained after taking action a in the current state, \(\gamma \) is the attenuation factor, \(s^{'} \) is the next state of s , and \(R_t+\gamma \max _{a^{\prime }} Q(s^{\prime }, a \mid \theta ^{\prime })\) is the Q value that needs to be optimized.

Because the daily stock price is often associated with the price in the previous days, the serialized update is selected as the update mechanism. After the epoch is randomly selected, it is deduced from the first day of the period T to the last day, then the network parameters are updated with the target network after every three episodes.

The investor’s goal is to trade stocks within the investment period T to maximize the cumulative return. For the convenience of subsequent calculations, the logarithm of assets at a certain time can be calculated in the form of compound interest as:

where \({A_0}\) represents the initial asset and \({A_t}\) represents the final asset.

Then the action space for stock trading is defined as buy, not buy. Not buy means being in a short position. The stock will be sold on the next day by default. In network training, in order to avoid the situation that only the same action is selected in the process, the \(\varepsilon \) -greedy algorithm is used. During the update process, the probability of choosing a random action decreases over time. In this paper, we have set the value of \({\varepsilon }\) to \({0.95^t}\) .

The equation is used at every decision point to select the optimal action, where there is a probability \({\varepsilon }\) of choosing a random action, and a probability of \({1-\varepsilon }\) of choosing the action that maximizes the Q-value, i.e., \(\underset{a}{\arg \max }\ Q(s,a)\) . During the update process, the probability of choosing a random action decreases over time. Based on the relationship between assets and daily returns, the reward function is set as follows:

where action value of 0 indicates not buying on the same day, action value of 1 indicates buying on the same day. At the same time, the reward value that is too large or too small in each transaction should be reduced to prevent excessive fluctuations [ 29 ].

The main purpose of introducing this formula is to smooth out the reward values at each time step, reduce fluctuations, and mitigate excessively high or low reward values. If \(\left| {{R_t}} \right| \le 0.01\) , the reward is set to reward * 0.2; if \(0.01 < \left| {{R_t}} \right| \le 0.02\) , the reward is set to reward * 0.7; and if the value of \(\left| {{R_t}} \right| \) is between 0.02 and 0.07, the reward remains unchanged.

4.4 Multi-head ProbSparse Self-attention mechanism

Stock prices are influenced by many factors, such as fundamental factors, technical factors, market factors, etc. The relationship between these factors is very complex, and it is difficult to effectively capture their interactions using traditional linear models. In 2021, Zhou et al. [ 35 ] proposed the informer model to improve the performance of the transformer model, with the proposal attention mechanism improving the traditional attention mechanism. The multi-head ProbSparse self-attention mechanism is capable of simultaneously focusing on multiple factors, capturing important features and their nonlinear relationships, thereby aiming to improve prediction accuracy. Therefore, multi-head ProbSparse self-attention can effectively capture the long-range dependencies between these factors, helping the model better understand the market state.

Structure of Multi-BiGRU with multi-head ProbSparse self-attention mechanism

Multi-head ProbSparse self-attention increases the model’s representational power by dividing the attention into multiple heads, with each head learning a different representation subspace. This allows the model to capture a more diverse range of information, which is helpful in generating more precise trading strategies. Additionally, ProbSparse attention introduces a probabilistic mechanism to sparsify attention weights, reducing the model’s focus on irrelevant information and thereby improving efficiency and robustness. This is particularly important when dealing with dynamically changing financial market data, as market conditions can shift rapidly, and multi-head ProbSparse self-attention can help the model adapt to these changes. Furthermore, the introduction of multi-head ProbSparse self-attention enhances the model’s generalization ability. By learning the intrinsic relationships between factors, the model can not only capture patterns in historical data but also make more reasonable predictions and decisions in the face of new market conditions. Therefore, we add the multi-head ProbSparse self-attention mechanism to DQN with Multi-BiGRU, allowing the model to handle complex financial market data more flexibly and thereby improve the performance of trading strategies, as shown in Fig. 6 .

After vectorizing the multiple factors, the multi-head ProbSparse self-attention mechanism maps the input vector to different subspaces, and each head performs self-attention calculations on different sub regions. In the self-attention mechanism of each head, we obtain three different matrices: query ( Q ), key ( K ), and value ( V ). Subsequently, when calculating the similarity scores between query and key, in traditional attention mechanisms, we need to use dot product to calculate the similarity between each \({q_i}\) in Q and each \({k_i}\) in K . However, in a large number of experiments, it can be found that for each \({q_i}\) , only a small portion of \({k_i}\) is closely related to it. So in order to filter out closely related \({q_i}\) and \({k_i}\) , we calculate the difference between each \({q_i}\) and a uniform distribution. If the difference is large, it indicates that the current \({q_i}\) is an ’active’ query that we need, otherwise it is a ’lazy’ query. The expression of the sparsity measure \(M({q_i},K)\) of the \({q_i}\) query is as follows:

where every \({q_i}\) , \({k_i}\) satisfies \({q_i}\in {{\mathbb R}^d},{k_i} \in {{\mathbb R}^d}\) , d is the feature dimension in the model and \({L_K}\) is the length of the key used for attention calculation. In order to reduce computational complexity, a portion of \({k_i}\) is sampled from the key for calculation, and the \({q_i}\) and \({k_i}\) with the highest inner product value are selected to calculate the difference from the uniform distribution. The equation is modified as follows:

After calculating the sparsity score, we select the largest n point product results, and use the mean vector as the reconstruction result for the remaining queries. The input and output lengths of the multi-head ProbSparse self-attention are both sequence length. Then the similarity between Q and K is obtained through dot product operation and normalized to [0,1] through softmax, and finally multiplied by the true value V to output the feature matrix. The mathematical representation of attention calculation is given below:

where \({\bar{Q}}\) is a sparse matrix of the same size as q , \({{d_k}}\) represents the dimension of Key and is used for weight scaling. The multi-head ProbSparse self-attention mechanism concatenates the output vectors of each head’s ProbSparse self-attention mechanism to obtain a complete output vector, further improving the efficiency and accuracy of obtaining feature importance weights in the sequence. The outputs from every head are concatenated as:

Where \({h_i}\) is the output of the i-th attention head. \({w_i^Q,w_i^K,w_i^V}\) are weight matrices that determine the degree of matching between queries and keys, as well as how to weight the values based on these matching degrees. In the Multi-BiGRU with multi-head ProbSparse self-attention mechanism, after the feature extraction, the attention mechanism translates the initial vector to the vector z and pass it as input into GRU network to obtain the corresponding Q value. When GRU network is learning, it will output the hidden state \(h_{t}\) and the memory state \(c_{t}\) , where the hidden state will be used as the input of the latter attention mechanism network, and the latter GRU network will consider all the previous hidden states and memory states together and output the corresponding Q value.

5 Experiment

5.1 experiment data.

Absolute technical indicators and relative technical indicators (ATI &RTI) are used as input data for the LightGBM model for turnover point classification. Multiple factors and sentiment factor are used as inputs for the reinforcement learning model which is trained to form trading strategies. Multiple factors are divided into valuation factors, growth factors, financial quality factors, leverage factors, size factors, momentum factors, volatility factors, turnover factors, shareholder factors, technical factors and emotional factors.

For the convenience of comparison, when comparing with non-temporal models, the selected samples are daily trading data of 40 stocks in the Shanghai and Shenzhen stock exchanges from January 1, 2011 to December 13, 2020. After data fusion and normalization, the data of first 1800 days are used as the training set and the remaining data as the test set. When comparing with temporal models, the selected samples are 30 stocks from both Chinese and US stock markets. The training period ranges from January 1, 2010, to January 1, 2018, and the test period ranges from January 1, 2018, to January 1, 2021. Except for the sentiment value, which requires crawling comments and conducting sentiment analysis, all factors for Chinese stock market stocks are implemented automatically, including technical factors (based on volume and price) and non-technical factors. These are obtained through interfaces in the ITendx quantitative platform using the DLL language (which is widely applicable across many platforms, similar to the PEL language used in the Pyramid platform). The same applies to the technical factors for stocks in the US market. Non-technical factors such as financial data are acquired through third-party libraries like akshare and yfinance, while the remaining few factors are manually retrieved from websites like Eastmoney and Trading Economics. The stock samples selected for comparison with non-temporal models and temporal models are presented in Tables 2 and 3 . Appendix C presents the multi factor (Table 4 ) data of 40 A-shares as of December 11, 2021.

5.2 Experiment settings

The operating system of the experimental machine is CentOS 8, the CPU is Intel Xeon Gold 5218, the graphics card is NVIDIA GeForce RTX 2080Ti, the code is written in Python language, and the main framework used is PyTorch.

When setting the various network parameters and structures, practical applications often rely on a trial-by-error approach for dynamic adjustment, but there are also certain empirical guidelines that can be followed. We Prevent overfitting through “early stopping” and “regularization,” and use the same number of neurons for all hidden layers. For the learning rate, we first set a very small learning rate, increase the learning rate after each epoch, and record the loss or accuracy for each epoch. The more epochs that are iterated, the more learning rates that are tested. Finally, we compare the loss or accuracy corresponding to different learning rates. The greedy action probability, similarly, decreases over the course of iteration, with the random action probability becoming smaller and smaller. The settings for the greedy action probability and sequence period are the values commonly adopted in most research in this field. Num_layers should not be too large, as it may pose risks such as overfitting and vanishing gradients.

In the DQN with Multi-BiGRU and multi-head ProbSparse self-attention, the following parameters are set: the deep Q network uses Multi-BiGRU, the num_layers is set to 2, the hidden_size is set to 128, the loss function is set to MSE, and the learning rate is 0.000025. In the multi-head ProbSparse self-attention mechanism, the num_heads is set to 8. The sequence period is 250 days, and the target network is used to update the network parameters after every 3 epochs. The probability of taking random actions in the greedy algorithm is \(0.95^{t}\) .

5.3 Experiment results

5.3.1 comparison of non-temporal models.

To verify the superiority of our model, we compared the method proposed by Wang et al. [ 29 ]. For the convenience of comparison, we selected 40 stocks from Shanghai and Shenzhen Stock Exchanges as experimental samples. The experimental results are shown in Table 5 .

From the experimental results in Table 5 , it can be seen that DQN+LightGBM with Multi-BiGRU and multi-head ProbSparse self-attention model outperforms DQN+LightGBM in stock tradings. Specifically, the returns achieved by our model are generally higher than those achieved by Buy and Hold. Among the 40 stock samples, our method achieved higher returns than DQN+LightGBM in 80% of the samples. For a small number of samples with poor performance such as 600053 and 000625, we found that most of their amplitudes are relatively large in the test set interval.

Boxplots of our model and non-temporal models’ performance

After completing the Friedman test for our model and the non-temporal models, as detailed in Table 6 , we observed statistically significant differences in the return outcomes between the model we developed and the four comparator baseline models. To further analyze the performance differences between each two models, we conducted a Nemenyi post-hoc analysis. The experimental results showed that the differences in outcomes (Table 7 ) between our model and DQN, DoubleDQN+LightGBM, and DuelingDQN+LightGBM are statistically significant, but the difference with DQN_CNN+LightGBM is not statistically significant. This is in line with Li’s research [ 14 ], which suggests that DQN performs better than DoubleDQN and DuelingDQN in terms of stock trading strategies. In order to more intuitively explore the distribution of experimental result data, as shown in Fig.  7 , we have plotted box plots for the experimental results, which reveal that our model provides superior performance in terms of return distribution and median values across a spectrum of individual stocks.

In summary, DQN+LightGBM with Multi-BiGRU and multi-head ProbSparse self-attention model is feasible on the sample dataset, and achieves higher returns than DQN+LightGBM model, it also has shown a significant difference in performance compared to the baseline, with a considerable advantage in terms of returns.

5.3.2 Comparison of temporal models

In order to compare the advantages of our model in algorithms of reinforcement learning with temporal neural network, we selected three models TFJ-DRL [ 13 ], PMMRL [ 18 ], and MSF-DRL [ 16 ] for comparative study. The experimental results are shown in Table 8 , where the profit rate of the comparative model is converted into percentage form.

As is shown in Table 8 , the samples include six different industries and are composed of stocks in Chinese and US markets. Our proposed method can achieve high returns within the test set and generally outperforms baseline models. Similarly, we can observe from a few underperforming stocks that most of them have significant amplitudes on the test set, such as 0700.HK, PTR, 300015.

Boxplots of temporal models’ performance

Following the Friedman test on the temporal models, Table  9 reveals that the model we introduced exhibits significant differences in terms of return outcomes compared to the four baseline models. To further analyze the performance differences between each two models, we conducted a Nemenyi post-hoc analysis. The experimental results showed that the differences in outcomes between our model and TFJ-DRL, PMMRL are statistically significant, but the difference (Table 10 ) with MSF-DRL is not statistically significant. In order to more intuitively explore the distribution of experimental result data, the boxplot in Fig. 8 illustrates that our model outperforms the others in terms of return distribution and median performance across various individual stocks.

5.3.3 Ablation studies

In this section, we conduct ablation studies on the 40 stocks from Shanghai Stock Exchange and Shenzhen Stock Exchange. The LightGBM model, Multi-BiGRU model and multi-head ProbSparse self-attention mechanism are investigated for the effectiveness of each module in more detail. Table 11 provides the ablation results of three configurations.

As is shown in Table 11 , we first evaluate the effectiveness of LightGBM turning point classification on the experimental results, and find that LightGBM turning point classification can reduce some wrong decisions in reinforcement learning to improve the returns achieved; we further evaluated the effectiveness of the multi-head ProbSparse self-attention mechanism on the results and found that it can better extract features from multiple factors to improve the returns of stock trading.

Boxplots of the performance in ablation studies

To better analyze the ablation experiment, we conducted a Friedman test on its results. As shown in Table 12 , there are significant differences between the proposed model and the model after removing modules. To further observe the performance (Tables 13 , 14 and 15 ) differences between each pair of models, we conducted a Nemenyi post-hoc test. The results reflected that each addition or removal (Tables 16 and 17 ) of a module had a significant impact on the outcome. From the box plot in Fig. 9 , it can be seen that the return distribution and median of our model steadily increased after adding LightGBM and multi-head ProbSparse self-attention, demonstrating the effectiveness of each module in the model.

6 Conclusion

In this paper, we propose a multi-factor stock trading strategy based on DQN with Multi-BiGRU and multi-head ProbSparse self-attention mechanism. In addition to considering the basic data of absolute and relative technical indicators, we also introduce factors such as valuation, financial quality, and sentiment values. These multi-factor data more comprehensively characterize the determinants of stock prices, and the Multi-BiGRU and multi-head ProbSparse self-attention mechanism are also capable of effectively extracting features from multi-factor time series data. In comparative experiments, we compare our model with the latest non-temporal and temporal models, with samples from different individual stocks in Chinese and US stock markets. Our model achieves better returns compared to other models. In ablation experiments, we demonstrate the necessity of each module in the proposed model. To further analyze the experimental results, we have drawn box plots and conducted Friedman test and Nemenyi post-hoc test and discussed the experimental results.

For future work, we consider the following two aspects: 1) there are still many uncertainties in stock data. How to reduce the uncertainty in stock data is a problem worth considering. 2) Risk control in stock investment is very important. Therefore, how to better balance returns and risks in stock trading strategies based on reinforcement learning is a topic worth researching.

Data Availability

Data will be made available on request.

Almahdi S, Yang SY (2017) An adaptive portfolio trading system: A risk-return portfolio optimization using recurrent reinforcement learning with expected maximum drawdown. Expert Syst Appl 87:267–279. https://doi.org/10.1016/j.eswa.2017.06.023

Aseeri AO (2023) Effective short-term forecasts of saudi stock price trends using technical indicators and large-scale multivariate time series. Peerj Comput Sci 9:e1205. https://doi.org/10.7717/peerj-cs.1205

Bollen J, Mao H, Zeng X (2011) Twitter mood predicts the stock market. J Comput Sci 2(1):1–8. https://doi.org/10.1016/j.jocs.2010.12.007

Article   Google Scholar  

Chakole JB, Kolhe MS, Mahapurush GD et al (2021) A q-learning agent for automated trading in equity stock markets. Expert Syst Appl 163:113761. https://doi.org/10.1016/j.eswa.2020.113761

Cho K, van Merriënboer B, Gulcehre C et al (2014) Learning phrase representations using RNN encoder–decoder for statistical machine translation. In: Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pp 1724–1734. https://doi.org/10.3115/v1/D14-1179

Cui C, Wang P, Li Y et al (2023) Mcvcsb: A new hybrid deep learning network for stock index prediction. Expert Syst Appl 232. https://doi.org/10.1016/j.eswa.2023.120902

Deng C, Huang Y, Hasan N et al (2022) Multi-step-ahead stock price index forecasting using long short-term memory model with multivariate empirical mode decomposition. Inf Sci 607:297–321. https://doi.org/10.1016/j.ins.2022.05.088

Fister D, Perc M, Jagric T (2021) Two robust long short-term memory frameworks for trading stocks. Appl Intell 51(10):7177–7195. https://doi.org/10.1007/s10489-021-02249-x

Hagenau M, Liebmann M, Neumann D (2013) Automated news reading: Stock price prediction based on financial news using context-capturing features. Decision Support Syst 55(3):685–697. https://doi.org/10.1016/j.dss.2013.02.006

Han H, Xie L, Chen S et al (2023) Stock trend prediction based on industry relationships driven hypergraph attention networks. Appl Intell. https://doi.org/10.1007/s10489-023-05035-z

Huang Z, Gong W, Duan J (2023) Tbdqn: A novel two-branch deep q-network for crude oil and natural gas futures trading. Appl Energy 347. https://doi.org/10.1016/j.apenergy.2023.121321

Huang Z, Li N, Mei W et al (2023) Algorithmic trading using combinational rule vector and deep reinforcement learning. Appl Soft Comput 147. https://doi.org/10.1016/j.asoc.2023.110802

Lei K, Zhang B, Li Y et al (2020) Time-driven feature-aware jointly deep reinforcement learning for financial signal representation and algorithmic trading. Expert Syst Appl 140:112872. https://doi.org/10.1016/j.eswa.2019.112872

Li Y, Ni P, Chang V (2020) Application of deep reinforcement learning in stock trading strategies and stock forecasting. Computing 102(6, SI):1305–1322. https://doi.org/10.1007/s00607-019-00773-w

Liu C, Yan J, Guo F et al (2022) Forecasting the market with machine learning algorithms: An application of NMC-BERT-LSTM-DQN-X algorithm in quantitative trading. ACM Trans Knowl Disc Data 16(4):1–22. https://doi.org/10.1145/3488378

Liu P, Zhang Y, Bao F et al (2023) Multi-type data fusion framework based on deep reinforcement learning for algorithmic trading. Appl Intell 53(2):1683–1706. https://doi.org/10.1007/s10489-022-03321-w

Liu W, Ge Y, Gu Y (2024) Multi-factor stock price prediction based on gan-trellisnet. Knowl Inf Syst. https://doi.org/10.1007/s10115-024-02085-8

Ma C, Zhang J, Liu J et al (2021) A parallel multi-module deep reinforcement learning algorithm for stock trading. Neurocomputing 449:290–302. https://doi.org/10.1016/j.neucom.2021.04.005

Ma G, Chen P, Liu Z et al (2022) The prediction of enterprise stock change trend by deep neural network model. Comput Intell Neurosci 2022:9. https://doi.org/10.1155/2022/9193055

Meng Q (2017) Lightgbm: A highly efficient gradient boosting decision tree. In: Neural information processing systems, pp 3149–3157

Mnih V, Kavukcuoglu K, Silver D et al (2013) Playing atari with deep reinforcement learning. arXiv:1312.5602

Park K, Shin H (2013) Stock price prediction based on a complex interrelation network of economic factors. Eng Appl Artif Intell 26(5–6):1550–1561. https://doi.org/10.1016/j.engappai.2013.01.009

Shi Y, Li W, Zhu L et al (2021) Stock trading rule discovery with double deep q-network. Appl Soft Comput 107:107320. https://doi.org/10.1016/j.asoc.2021.107320

Soleymani F, Paquet E (2020) Financial portfolio optimization with online deep reinforcement learning and restricted stacked autoencoder-deepbreath. Expert Syst Appl 156. https://doi.org/10.1016/j.eswa.2020.113456

Staffini A (2022) Stock price forecasting by a deep convolutional generative adversarial network. Front Artif Intell 5. https://doi.org/10.3389/frai.2022.837596

Taghian M, Asadi A, Safabakhsh R (2022) Learning financial asset-specific trading rules via deep reinforcement learning. Expert Syst Appl 195. https://doi.org/10.1016/j.eswa.2022.116523

Takara LdA, Santos AAP, Mariani VC et al (2024) Deep reinforcement learning applied to a sparse-reward trading environment with intraday data. Expert Syst Appl 238(C). https://doi.org/10.1016/j.eswa.2023.121897

Touzani Y, Douzi K (2021) An LSTM and GRU based trading strategy adapted to the Moroccan market. J Big Data 8(1):126. https://doi.org/10.1186/s40537-021-00512-z

Wang J, Jing F, He M (2023) Stock trading strategy of reinforcement learning driven by turning point classification. Neural Process Lett 55(3, SI):3489–3508. https://doi.org/10.1007/s11063-022-11019-w

Watkins CJCH (1989) Learning from delayed rewards. PhD thesis, Cambridge University

Wu ME, Syu JH, Lin JCW et al (2021) Portfolio management system in equity market neutral using reinforcement learning. Appl Intell 51(11):8119–8131. https://doi.org/10.1007/s10489-021-02262-0

Wu X, Chen H, Wang J et al (2020) Adaptive stock trading strategies with deep reinforcement learning methods. Inf Sci 538:142–158. https://doi.org/10.1016/j.ins.2020.05.066

Yang Z, Zhao T, Wang S et al (2024) Mdf-dmc: A stock prediction model combining multi-view stock data features with dynamic market correlation information. Expert Syst Appl 238(E). https://doi.org/10.1016/j.eswa.2023.122134

Yu X, Li D (2021) Important trading point prediction using a hybrid convolutional recurrent neural network. Appl Sci-Basel 11(9):3984. https://doi.org/10.3390/app11093984

Zhou H, Zhang S, Peng J et al (2021) Informer: Beyond efficient transformer for long sequence time-series forecasting. In: Proceedings of AAAI, pp 11106–11115

Download references

Acknowledgements

This work is supported by the National Natural Science Foundation of China (62071240), the Innovation Program for Quantum Science and Technology (2021ZD0302901), the Natural Science Foundation of Jiangsu Province (BK20220804, BK20231142), and the Priority Academic Program Development of Jiangsu Higher Education Institu-tions (PAPD).

Author information

Authors and affiliations.

School of Software, Nanjing University of Information Science and Technology, Nanjing, 210044, People’s Republic of China

Wenjie Liu, Yuchen Gu & Yebo Ge

Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology (CICAEET), Nanjing, 210044, People’s Republic of China

Jiangsu Province Engineering Research Center of Advanced Computing and Intelligent Services, Nanjing University of Information Science and Technology, Nanjing, 210044, People’s Republic of China

You can also search for this author in PubMed   Google Scholar

Contributions

WenJie Liu: Conceptualization, Data curation, Formal analysis, Methodology, Project administration, Investigation, Software, Supervision, Funding acquisition, Writing - original draft, Writing - review & editing. YuChen Gu: Methodology, Data curation, Formal analysis, Software, Investigation, Visualization, Resources, Writing - original draft. YeBo Ge: Formal analysis, Investigation, Validation, Writing - review & editing. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Wenjie Liu .

Ethics declarations

Competing interest.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Ethical and informed consent for data used

This article does not contain any studies with human participants or animals performed by any of the authors.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix A Glossary

The following table shows the stock terms in the article and their corresponding definitions.

Appendix B Multiple factors

The following tables show descriptions of absolute technical indicators, relative technical indicators and complete multiple factors.

Appendix C Multi-factor data of 40 A-shares

The following table shows the multi-factor data of 40 A-shares on December 11, 2020.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Liu, W., Gu, Y. & Ge, Y. Multi-factor stock trading strategy based on DQN with multi-BiGRU and multi-head ProbSparse self-attention. Appl Intell (2024). https://doi.org/10.1007/s10489-024-05463-5

Download citation

Accepted : 11 April 2024

Published : 22 April 2024

DOI : https://doi.org/10.1007/s10489-024-05463-5

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Deep reinforcement learning
• Multi-factor
• Multi-BiGRU
• Multi-head ProbSparse self-attention
• Stock trading strategy
• Find a journal
• Publish with us
• Track your research