A meta-analysis of effects of blended learning on performance, attitude, achievement, and engagement in different countries

Although the new century has been witnessing increasing popularity of blended learning especially during this special pandemic time, few studies have summarized the effectiveness of blended learning in different countries. This meta-analysis summarizes previous studies on blended learning effectiveness in different countries in terms of students’ performance, students’ attitudes towards blended learning, learning achievement, and student engagement in different countries. Through the meta-analysis via Stata/MP 14.0, it is concluded that blended learning could improve performance, attitude, and achievement in most countries. However, in both China and the USA, blended learning could not signi�cantly improve student engagement in academic activities. No signi�cant differences were revealed in student performance in the USA between blended and non-blended learning. Future research could extend the research into blended learning to more countries and areas across the world.


Introduction
This special pandemic time has been witnessing the popularity of blended learning approaches.However, very few studies have summarized blended learning effectiveness in different countries.It is thus meaningful and necessary to examine the effectiveness of blended learning across the world especially during this special time.

Performance
Most studies positively reported blended learning performances.Blended learning, outperforming full online learning in the aspects of motivation, attitudes, and satisfaction, could improve nurses' clinical knowledge compared with the traditional learning approach in the UK (McCutcheon, O'Halloran, & Lohan, 2018).Compared with traditional face-to-face learning, blended learning could optimize the learning exibility in terms of time and space, leading to stable learning performance of undergraduates in The Zurich University in Germany (Mueller, Mildenberger, & Lübcke, 2020).It was revealed that both classroom and online learning could enhance American students' learning performance, but the blended learning brought about the largest gain in performance in the USA (Hill, Chidambaram, & Summers, 2017).Blended learning could lead to signi cantly higher learning performance than e-learning, while the ipped classroom could improve intrinsic motivation and self-e cacy in Can Tho University in Vietnam (Thai & Valcke, 2017).
Numerous studies reported that blended learning was bene cial to language pro ciency improvements.
Blended learning could greatly improve the reading abilities of children in a kindergarten in the USA (Macaruso, Wilkes, & Prescott, 2020).Blended instruction could greatly improve students' English writing abilities, e.g.content relevance, content su ciency, organization structure, and language expression in Ankang College, Shanxi China (Zhou, 2018).Blended learning could improve students' English listening and speaking and critical thinking skills, e.g.analysis, inference, evaluation, induction, and deduction in China (Yang, Chuang, Li, & Tseng, 2013).Blended learning could enable Chinese college students to extensively practice with exible time and space, greatly improving their English reading skills (Yang, 2012).
Blended learning could also enhance high-order abilities such as communication, problem-solving, and reasoning skills.Blending a class video blog into face-to-face instruction could improve language oral pro ciency but failed to greatly improve the voluntariness to communication using the target language in China (Liu, 2016).Blended learning could effectively facilitate communication skills and improve learning outcomes of nursing tertiary students in Singapore (Shorey, Kowitlawakul, Kamala Devi, Chen, Soong, & Ang, 2018).In the blended learning, Chinese students could discuss with peers, propose meaningful ideas, mutually learn and share, improve group work skills, enhance self-perception, and facilitate reasoning skills (Monteiro & Morrison, 2014).Blended learning could enhance acute stroke patients' competences, e.g.recognition and management in the USA (Lee Gordon, Issenberg, Gordon, Lacombe, Mcgaghie, & Petrusa, 2005).

Attitude
Most of the blended learning participants held positive attitudes towards blended learning effectiveness.
Blended learning, conducive to students' positive attitude and satisfaction, could improve English listening skills and enhance vocabulary acquisition among junior middle school students in China (Jia, Chen, Ding, & Ruan, 2012).Chinese 11th graders held signi cantly more positive attitudes towards blended learning than traditional learning (Chang, Shu, Liang, Tseng, & Hsu, 2014).Singaporean nursing college students had greatly positive attitude towards blended learning, as well as communication skills in the blended context (Shorey, Kowitlawakul, Kamala Devi, Chen, Soong, & Ang, 2018).The blended model in active learning classrooms obtained positive evaluation and students held improved attitudes towards physics courses in North Carolina State University in the USA (Beichner, Saul, Abbott, Morse, Deardorff, & Allain et al., 2007).Blended learning could improve nursing students' motivation, satisfaction, and attitude in clinical supervision skills compared with online-only learning in China (Chang, Shu, Liang, Tseng, & Hsu, 2014).

Achievement
Many studies reported that blended learning could contribute to higher learning achievements than traditional approaches.Blended learning could lead to signi cantly higher academic achievements than traditional face-to-face learning in Canada (Bazelais & Doleck, 2018).Online learning activities could improve students' academic achievements among undergraduate students in University of Granada in Spain, where in uencing factors included attendance rate and students' backgrounds rather than the time they spent on learning (López-Pérez, Pérez-López, Rodríguez-Ariza, & Argente-Linares, 2013).Blended learning via information and communication technologies could signi cantly improve learning achievements of mechanical couplings in engineering in Spain (Cortizo, Rodriguez, Vijande, Sierra, & Noriega, 2010).A blended and ipped pedagogical approach could improve learning achievements and learning environment and raise the e ciency of space use in the USA (Baepler, Walker, & Driessen, 2014).

Engagement
Most previous studies reported that blended learning could improve learning engagement.Blended learning, encouraging students to engage in learning even after class, could lead to a signi cantly higher frequency and level of engagement than the traditional learning in Spain (Pérez-Marín & Pascual-Nieto, 2011).In the technology-oriented blended learning, Chinese freshmen used to spend more time on inclass discussion and writing tasks than the e ciency-oriented group.The interaction was considered an important factor in uencing blended learning effectiveness among Chinese freshmen (Yen & Lee, 2011).Undergraduates at Point Loma Nazarene University in the USA spent signi cantly more time learning in a blended instruction model than in the traditional instruction model (Botts, Carter, & Crockett, 2018).
Blended learning could improve Chinese students' engagement by increasing their learning e ciency and effectiveness (Monteiro & Morrison, 2014).

Contradictory ndings
However, there were contradictory ndings about the effect of blended learning in different countries in terms of achievement.It was reported that blended learning could not signi cantly improve students' achievements in China although they themselves believed so (Chang, Shu, Liang, Tseng, & Hsu, 2014).No signi cant differences in Fashion learning achievements were revealed in Hong Kong China between blended learning and traditional face-to-face learning (Yick, Yip, Au, Lai, & Yu, 2019).Blended learning did not indicate any effect on learning outcomes of economics courses in an American university (Olitsky & Cosgrove, 2014).While blended learning could signi cantly improve self-assessment of students' knowledge gains, it could not greatly improve their actual learning achievements among Chinese Vocational High School Students (Chang, Shu, Liang, Tseng, & Hsu, 2014).
Contradictory ndings were also found regarding the effect of blended learning on attitude, performance, and engagement.There are no signi cant differences in attitudes towards blended or traditional learning, which might be subject to academic evaluation, teaching experience, and computer skills in the United Arab Emirates (Al-Qatawneh, Eltahir, & Alsalhi, 2020).Different models of blended learning could lead to various performance levels.E ciency-oriented blended learning could signi cantly improve problemsolving performance of Chinese freshmen than hybrid-oriented and the technology-oriented groups (Yen & Lee, 2011).No signi cant difference in learning outcomes was found between blended and traditional instructions in the upper-division quantitative literacy course at Point Loma Nazarene University in the USA (Botts, Carter, & Crockett, 2018).In blended learning, students were less engaged in problem-solving issues and students in Point Loma Nazarene University in the USA spent less time in course learning, followed by reduced seat time (Botts, Carter, & Crockett, 2018).No signi cant differences in self-e cacy and knowledge score were revealed between blended and non-blended methods although students positively perceived blended learning at an undergraduate university in Alberta, Canada (Berga, Vadnais, Nelson, Johnston, & Olaiya, 2021).

Research questions
Considering the inconsistent ndings regarding the effect of blended learning on performance, student attitude, achievements, and engagement in different countries, we propose four research questions, i.e.
(1) Could blended learning positively in uence student performance in different countries?(2) Could blended learning positively in uence student attitude in different countries?(3) Could blended learning positively in uence learning achievement in different countries?(4) Could blended learning positively in uence student engagement in different countries?

Methods
This meta-analysis was implemented according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Moher, Liberati, Tetzlaff, Altman, & The PRISMA Group, 2009).The registration of the meta-analytical protocol was waived by the author's academic review board due to the characteristics of the study.

Eligibility criteria
The studies will be considered eligible and included if they (1) focus on the effect of blended learning on performance, student attitude, achievements, and engagement in different countries; (2) are highly evaluated using University of West England Framework for Critically Appraising Research Articles (Appendix A) (Moule et al., 2003); (3) can provide enough data for a meta-analysis; (4) divide the participants into both control and experimental groups for a comparative analysis between blended learning and non-blended learning; and (5) are written in the standard English language.
The studies will be considered ineligible and excluded if they (1) focus on blended learning technologies themselves rather than blended learning effect; (2) cannot provide enough data for a meta-analysis even after we correspond with the authors; (3) are not written in English; or (4) they are poorly evaluated using University of West England Framework for Critically Appraising Research Articles (Appendix A) (Moule et al., 2003).

Data sources and search strategy
Based on the PRISMA ow (Fig. 1), we conducted the inclusion and exclusion process.To maximize the number of data included, we searched the databases from their inception until February 26, 2021 without time limitation.We entered keywords and index terms, e.g.blended learning, performance, attitude, achievements, and engagement, different countries, into different databases according to their speci c syntactical rules.We obtained 12098 results by searching four online databases, i.e.Elsevier ScienceDirect, Taylor & Francis Group, EBSCOhost, and Springer.Then we entered the results into ENDNOTE X8 (Thomson Reuters, New York, USA) to remove those duplicated.Then we invited two researchers to double-check whether or not the results are related to the study by screening the titles and abstracts.After this, they conducted the evaluation of eligibility of the results.
Finally, two researchers met to decide on the included studies for the meta-analysis.They discussed different selected studies and negotiated to address the disputes.Those selected by both of them were directly included in the meta-analysis.A third reviewer will be invited to nally determine the nally selected studies in case two researchers cannot reach an agreement on the inclusion of any study.

Evaluation of included studies
We evaluated the full texts via University of West England Framework for Critically Appraising Research Articles (Appendix A) (Moule et al., 2003).This framework evaluates the research articles based on ve sections, i.e. the introduction, the methods, ethics, the results/ ndings, and the conclusions.Each section has detailed criteria for evaluation.For example, for the section of introduction, we examine whether there is a clear statement about the topic being investigated and whether there is a clear rationale for the research.For the conclusion section, we examine whether the implications for further research are acknowledged, whether areas for further research are identi ed, and whether further recommendations are made for practice that come from the results.For the method section, we use different criteria for different methods, e.g.qualitative or quantitative research.We also use speci c criteria to evaluate data collection and analysis.We nally included 26 results for the meta-analysis (Table 1).The inter-rater consistency reaches a satisfactory level (Cohen's kappa coe cient = 0.83).This indicates that two researchers mostly selected the same studies or generally reached an agreement on most of the selected studies.

Data extraction
As shown in Table 1, two researchers extracted speci c information such as author, publication year, and the source of the literature.We also collected enough data for the meta-analysis such as means, standard deviations, and numbers of participants for both control and experimental groups.For convenience of analysis, we classi ed the ndings into performance, attitude, achievements, and engagement, followed by the countries where the studies were conducted.The selected were implemented in various countries across the world such as China, the United Arab Emirates, Canada, the USA, Spain, Germany, Singapore, and Vietnam.We will compare different effects of blended learning in these countries.Similarly, both researchers would meet up to discuss different results of data extraction and a third reviewer would be invited to decide the nal data if any disagreement occurred between two researchers.The inter-rater consistency also reaches a satisfactory level (Cohen's kappa coe cient = 0.81).

Statistical analysis
We meta-analytically examined the data using Stata/MP 14.0.After entering data such as numbers of participants, means, and standard deviations for both groups into Stata/MP 14.0, forest plots will be drawn.We calculated standardized mean difference (SMD or Cohen d) (Cohen, 1988) indicating the effect sizes, weight indicating the degree of the in uence on pooled results, and 95% con dence interval indicating the study reliability.Cohen d is produced through dividing the mean difference between both groups by the pooled standard deviation of both groups (Sedgwick & Marston, 2013).The formula is:  (Sawilowsky, 2009).
To determine whether a random-effect or a xed-effect model could be adopted, we also tested the heterogeneity of the effect sizes using I 2 and p values.The formula to calculate I 2 is: , where Q indicates the Chi-squared statistics and df means the degree of freedom (Higgins & Thompson, 2002;Higgins, Thompson, Decks, & Altman, 2003).This indicates the degree of percentage of the variability in effect sizes caused by heterogeneity or random errors.According to Higgins & Green (2021), the heterogeneity will be considered not important in case I 2 ranges from 0-40%, moderate in case I 2 ranges from 30-60%, substantial in case I 2 ranges from 50-90%, and considerable in case I 2 ranges from 75-100%.Generally, if I 2 is larger than 50% (p < .05),we will adopt a random-effect model to conduct the meta-analysis, and if I 2 is smaller than 50% (p > .05),we will use a xed-effect model to run the meta-analysis.The in uence analysis program will be used to run the sensitivity analysis.Both Begg's (Begg & Mazumdar, 1994) and Egger's (Egger, Smith, Schneider, & Minder, 1997) tests will be used to test the publication bias.

Tests of publication bias
To test the publication bias, we rstly entered data, e.g.means, standard deviation, and numbers of participants across both groups, into Stata/MP 14.0 to run the meta-analysis.Then, we obtained effect sizes (ES) and standard errors of effect sizes (seES) for the test of publication bias.We tested the publication bias by entering "ES, seES" into Stata/MP 14.0, leading to a funnel plot (Fig. 2) and related data.A dot indicates an individual study, and the middle line is the no-effect line.If the dots are symmetrically distributed along both sides of the no-effect line, there will be an absence of publication bias.On the contrary, the asymmetrical distribution indicates the presence of publication bias.As shown in Fig. 2, it is hard to conclude that the dots are symmetrically distributed, indicating the presence of publication bias.Both Begg's (Q = 1016, S.D. = 381.89,z = 2.66, p = 0.008) and Egger's tests (Coe cient = 1.55,S.E.= .48,t = 3.25, p = 0.002, 95%CI = .60~ 2.48) also indicate the presence of publication bias.

A sensitivity analysis
We conducted a sensitivity analysis to test the reliability and stability of the obtained effect sizes using the program "metan-based in uence analysis".To retrieve the result, we entered the data such as means, standard deviations, and numbers of both groups into Stata/MP 14.0.We adopted a random-effect model to conduct the sensitivity analysis due to the high degree of percentage of variability caused by heterogeneity (Q = 1053.01,I 2 = 89.7%,z = 8.88, p < .01).
Unstable ES estimates often lead to skewed distribution and are frequently located beyond the lower and upper bounds of 95% con dence intervals (Borenstein, Hedges, Higgins & Rothstein, 2009).It is thus a must to identify whether there is any estimate located beyond the scope of 95% con dence intervals (Borenstein, Hedges, Higgins & Rothstein, 2009).As shown in Fig. 3, a dot indicates an estimated effect size of an individual study.All the effect sizes are located within the low and upper bounds of 95% con dence intervals.This indicates that there are no unstable ES estimates.We, therefore, conclude that the meta-analysis results are stable.

Could blended learning positively in uence student performance in different countries?
To determine student performance in blended and non-blended learning modes in different countries, we retrieved 27 effect sizes from different countries, where 18 effect sizes sourced from China, 8 from the USA, and 1 from Vietnam.We failed to obtain an effect size from a study (Yang, Chuang, Li, & Tseng, 2013) because one of the standard deviation values is zero.We obtained meta-analytical data and a forest plot (Fig. 4) after entering means, standard deviations and, numbers across both groups into Stata/MP 14.0 to run the meta-analysis by the variable country.
As shown in Fig. 4, the diamonds at the bottom indicate the pooled results.In the left-most column are displayed the author names and publication years, followed by a middle line with numerous boxes.The middle line is referred to as a no-effect line because if a diamond crosses it, the result will be considered insigni cant.A box, integrated with a horizontal line and a dot, indicates an individual study.The length of the horizontal line is negatively related to the reliability of the study.The dot indicates the SMD.On the right are displayed the statistics of SMDs (Cohen d) and 95% con dence intervals after them.The rightmost column shows the weights indicating the in uence of effect sizes on the pooled result.
We adopted a random-effect model to run the meta-analysis of the data sourcing from China (I 2 = 91.2%,p < .01), the USA (I 2 = 90.4%,p < .01)and Vietnam (a single study) due to a generally high degree of percentage of variability caused by heterogeneity (I 2 = 92.9%,p < .01).
As for the meta-analysis of data sourcing from China and Vietnam, the diamonds are located to the right of the no-effect line.This indicates that student performances in the blended learning context in China (d = 0.77, 95%CI = 0.44 ~ 1.10, z = 4.59, p < .01)and Vietnam (d = 0.66, 95%CI = 0.06 ~ 1.27, z = 2.14, p = 0.032) are signi cantly higher than the non-blended.However, the diamond retrieved from the data sourcing from the USA crossed the no-effect line, indicating that student performance in the blended learning context in the USA (d = -0.02,95%CI = -0.27~ 0.23, z = 0.19, p = 0.853) is not signi cantly higher than the non-blended.The overall results indicate that the blended learning can lead to signi cantly (d = 0.50, 95%CI = 0.27 ~ 0.74, z = 4.24, p < .01)higher student performance than the non-blended since the diamond is located to the right of the no-effect line.In general, we believe that blended learning could positively in uence student performance in different countries.

Could blended learning positively in uence student attitude in different countries?
To determine the differences in student attitudes between blended and non-blended learning in different countries, we obtained totally 11 effect sizes from the studies sourcing from the United Arab Emirates, China, Singapore, Vietnam, the UK, and the USA.We adopted a random-effect model to conduct the metaanalysis due to the high degree of percentage of variability of the effects sizes sourcing from different countries caused by heterogeneity (I 2 = 76.9%,p < .01).

Could blended learning positively in uence learning achievement in different countries?
To identify students' achievements of blended learning in different countries, we extracted 57 effect sizes, where 2 of them sourced from Canada, 13 from China, 22 from Germany, 3 from Spain, 1 from the United Arab Emirates, 1 from the UK, 1 from Vietnam, 14 from the USA.We adopted a random-effect model to implement the meta-analysis due to the high degree of percentage of variability caused by heterogeneity (I 2 = 87.4%,p < .01).We entered means, standard deviations, and numbers of participants across both groups into Stata/MP 14.0, then we obtained a forest plot after running the meta-analytical program by the variable country (Fig. 6).
The pooled diamond at the bottom is located to the right of the no-effect line without crossing it.We thus conclude that the students' overall achievement in the blended learning context is signi cantly larger than that in the non-blended learning context (d = 0.30, 95%CI = 0.21 ~ 0.40, z = 6.24, p < .).We, therefore, believe that blended learning could positively in uence learning achievement in different countries.

Could blended learning positively in uence student engagement in different countries?
To identify whether the blended approach could improve student engagement in learning, we extracted 14 effect sizes, where 3 of them sourced from the USA, and 11 from China.We adopted a random-effect model to conduct the meta-analysis due to a high degree of percentage of variability of effect sizes caused by heterogeneity (I 2 = 89.5%,p < .01).After entering means, standard deviations, and numbers of participants of both groups into Stata/MP 14.0, we obtain a forest plot (Fig. 7) from the meta-analysis by the variable country.As shown in Fig. 7, the diamonds obtained from the meta-analysis of data sourcing from both China and the USA cross the no-effect middle line.The diamond for the overall result also crosses the no-effect middle line.We thus conclude that there are signi cant differences in student engagement between blended and non-blended learning in both China (d = 0.14, 95%CI = -0.06~ 0.34, z = 1.38, p = 0.169), the USA (d = 0.51, 95%CI = -0.35~ 1.38, z = 1.16, p = 0.245), and the overall results (d = 0.23, 95%CI = -0.09~ 0.55, z = 1.42, p = 0.156).Therefore, we believe that blended learning could not positively in uence student engagement in different countries.

Discussion
This meta-analysis nds that blended learning could positively in uence student performance, student attitude, and learning achievement rather than student engagement in different countries.Most of previous studies found positive in uence of blended learning on performance, attitude, and achievement (e.g.Yen & Lee, 2011;Chang, Shu, Liang, Tseng, & Hsu, 2014), while some revealed negative in uence of blended learning on the engagement (e.g.Botts, Carter, & Crockett, 2018).Numerous factors account for the ndings.
Student performance tends to be enhanced in the blended learning context.With the blended learning method, students learn both in the face-to-face physical classroom and online with information technologies.In the classroom, they can interact with peers and teachers directly for academic issues.
They are also under the supervision of the teacher, encouraging them to be involved in learning activities.
They are often required to focus on the learning contents because the teacher tends to ask them to answer questions.They have frequent eye contacts with the teacher, which urges them to follow the teaching progress.If they nd anything hard to comprehend, they can ask the teacher by raising their hands.After class, they can also keep pace with learning and teaching progress conveniently using online communication technologies.They can log into the Internet and have access to a learning platform to acquire knowledge and share abundant learning resources.They neither need to carry heavy bags and books nor travel a long distance to the classroom.They can learn through the platform whenever and wherever they feel comfortable (Yu & Yi, 2020).This convenient blended mode undoubtedly improves their performance.
Students also tend to hold positive attitudes towards blended learning.The blended learning method brings great convenience to them.They can carry a light portable smart phone everywhere to acquire knowledge and learn with them in their hands or even on a bus.The small device can carry a sea of learning materials.Otherwise, they have to carry tons of books to learn page by page in a physical classroom.They can also interact through the online platform with their peers to address di cult problems and cultivate a favorable learning environment (Yu, Zhu, Yang, & Chen, 2019).Blended learning makes full use of the power of the Internet, combining the online learning with classroom learning.In this way, there evolves an autonomous learning approach, consisting of real-time and non real-time, synchronous and asynchronous learning, opinion sharing, collaborative learning, and group learning based on the concept of cooperation.The formal instruction and informal learning will be seamlessly linked, de nitely improving students' attitudes toward blended learning.
Students tend to achieve great academic success in blended learning.Blended leaning requires the instructors to put the learning materials on the Internet or on a learning platform, and the learners can browse these materials at any time according to their needs.Contact information of experts and instructors is openly accessible on the Internet.If learners encounter problems or want to further explore the knowledge, they can contact relevant experts or instructors at any time.Face-to-face communication can also be used in the physical classroom.Online courses play a very important role in the transmission of learning materials and contents, which facilitates learners or trainees to communicate with each other (Yu, & Wang, 2016).This is conducive to strengthening their in-depth understanding of the learning contents.Through the knowledge constructed by learners themselves, blended learning can creatively form their own thoughts and share with others through the Internet.
Nevertheless, in the blended learning context, students' engagement may be weakened due to various factors.Some online learning platforms have numerous problems, such as unstable running system environment, slow computer operation speed, easy to jam in online processing, rigid interface design, and lack of humanization.The unreasonable design of menus may cause an abomination of learners.If the platform is not compatible with some software, the system may be unstable or even suspend the learning procedure.If the Internet is too slow, students will not be able to open multiple windows at the same time, which may frustrate their learning enthusiasm, weaken their computer self-e cacy (Sun & Rueda, 2012), and they may nally abandon the blended learning approach.
Various backgrounds of learners may cause different feedback to blended learning.As learners come from all over the world, they, subject to geographical, family, school, and other in uencing factors, may have various responses to online learning technologies.It is thus necessary for an educational institute to carry out a training program to improve students' skills of online technology use (Bernard, Borokhovski, Schmid, Tamim, & Abrami, 2014).They can also train students using a recorded video where the operation instruction can be detailed.An educational department can regularly implement training programs of blended learning to bridge the gap of technology skills among students.In this way, students with different backgrounds can adapt themselves to the blended learning environment.
To improve the engagement in blended learning, it is important to design appropriate curriculum contents (Vaughan, 2007).Teachers or instructors need to integrate the advantages of online learning into o ine classroom teaching, design the curriculum based on the needs of learners, provide visual and aural learning stimuli, enhance students' self-e cacy, trigger their learning interest, enhance their learning engagement, and nally raise their level of knowledge.With strong self-e cacy or satisfaction, students may feel voluntary to receive blended instruction and keep regular engagement in blended learning programs.To improve students' engagement, MOOCs could be used as a form to blend face-to-face courses (de Moura, de Souza, & Viana, 2021).

Major ndings
This meta-analysis mainly examined the differences in performance, attitude, achievement, and performance between blended and non-blended learning approaches in different countries.It is revealed that blended learning could improve performance, attitude, and achievement in most countries.However, in both China and the USA, blended learning could not signi cantly improve student engagement in academic activities.No signi cant differences were revealed in student performance in the USA between blended and non-blended learning.

Limitations
There are several limitations to this study.Firstly, the meta-analysis cannot include all the publications and non-published works due to the limitation of library sources.Secondly, Both Begg's and Egger's tests indicate the presence of publication bias.Thirdly, we could not completely retrieve the identi ed research due to various reasons.

Future research directions
Whether in blended or non-blended learning contexts, teachers need to speci cally instruct students to collaborate and practice, where students can learn the bene ts and challenges of blended or non-blended learning and improve their learning effectiveness (Monteiro & Morrison, 2014).Although the new century has been witnessing the increasing popularity of blended learning especially in this special pandemic time, few studies have been committed to the effectiveness of blended learning in different countries.
Future research could extend the research into blended learning to more countries and areas across the world.

Declaration
Competing interests: The author declares no competing interests.

Figure 7 .
Figure 7.A forest plot of students' engagement in different countries Figures

Figure 5 A
Figure 5

Table 1
The included studies for the meta-analysis where M1 indicates the mean of the control group) and M2 indicates the mean of the experimental group.The effect size will be deemed very small if d approximates 0.1, small if d approximates 0.2, medium if d approximates 0.5, large if d approximates 0.8, very large if d approximates 1.2, huge if d approximates 2.0