A meta-analysis of gender differences in e-learning outcomes

The new century has been witnessing a rapid development of information technologies, along with which e-learning has been increasingly popularized especially in this special pandemic time. This study, including 20 high-quality publications, meta-analytically examined gender differences in e-learning outcomes, e.g. e-learners’ self-ecacy, satisfaction, motivation, attitude, and performance, across the world. The study concludes that there are generally no signi�cant gender differences in e-learning outcomes except in a few countries. For example, females signi�cantly outperformed males in Spain and the UK. In Austria, India, and mixed countries (Chile and Spain), females hold signi�cantly more positive attitudes towards e-learning than males. In the USA, females present signi�cantly higher self-ecacy than males. Future research into the gender issue in e-learning across the world may adopt cross-disciplinary research methods except for a meta-analysis.


Introduction
With rapid development of science and technology, the new century has been witnessing growing selfe cacy, satisfaction, motivation, attitude, and performance among e-learners (Thompson, Meriac, & Cope, 2002).

Self-e cacy
Previous studies reported signi cant differences in e-learning self-e cacy (e.g.Chen & Tsai, 2007).Selfe cacy in e-learning, positively in uencing e-learning effectiveness (Hsu & Chiu, 2004), is operationally de ned as the individual evaluation of the e-learning experience and the individual ability to complete a given e-learning task (Torkzadeh & Van Dyke, 2002).Presence of males can lead to signi cantly higher self-e cacy than females (Baylor & Kim, 2004).Learners with higher self-e cacy could be able to obtain more knowledge by focusing on online resources, perform better by spending more time and be more motivated to engage in e-learning than those with lower self-e cacy (Pituch & Lee, 2006).Females, with lower self-e cacy, tend to be more subject to unskillful use of e-learning technology than males in China (Ong & Lai, 2006).Compared with males, females in China tend to increase their self-e cacy dependent on their family support, which indicates that e-learning is closely related to social contexts of genders rather than sex itself (Chu, 2010).

Motivation and satisfaction
Previous studies have provided contradictory ndings regarding gender differences in e-learning satisfaction and motivational gender difference has generally not been revealed in Malaysia (Marimuthu, Chone, Heng, Nah, & Fen, 2013).No signi cant gender differences have been revealed in e-learning motivation and satisfaction although e-learning through the mobile platform -Moodle may positively in uence e-learning satisfaction and motivation for both males and females in Spain and the UK (Cuadrado-García, Ruiz-Molina, & Montoro-Pons, 2010).No signi cant effect of gender and age on elearning readiness or satisfaction has been revealed in Hong Kong, China (So & Swatman, 2010).There is no signi cant gender difference in e-learning motivation (Yukselturk & Bulut, 2009).There are also other studies reporting no signi cant gender differences in satisfaction (e.g.Ramírez-Correa, Arenas-Gaitán, & Rondán-Cataluña, 2015) with and attitudes towards the e-learning approach (e.g.Cuadrado et al., 2010;Hung, Chou, Chen, & Own, 2010) although Hong (2002) argues that gender plays an important role in elearners' satisfaction.
Nevertheless, it is reported that females, planning learning schedules and interacting with instructors more effectively, tend to be more satis ed with e-learning courses than males among mixed participants in Spain and the UK (González-Gómez, Guardiola, Martín Rodríguez, & Montero Alonso, 2012).Females consider e-learning effective and are thus more satis ed with it than males (Hu & Hui, 2011) although elearning motivation of females is signi cantly lower than that of males (Hu & Hui, 2011).Reverse ndings were found by Lu and Chiou (2010) who reported that males were more satis ed with e-learning than females.Social presence in e-learning could improve learners' motivation and satisfaction (Thayalan, Shanthi, & Paridi, 2012).Males feel signi cantly more enjoyable and satis ed with e-learning via video models (Hoogerheide, Loyens, & Van Gog, 2016).

Performance
Previous studies have arrived at inconsistent conclusions regarding the gender differences in e-learning performance (Price, 2006;Marimuthu, Chone, Heng, Nah, & Fen, 2013).No gender differences were revealed in e-learning performance (Chen & Tsai, 2007).Gender has also been considered an insigni cant in uencing factor in e-learning performance (Yukselturk & Bulut, 2009).However, gender differences were found in use of technology, e-instruction, technology skillfulness, and information literacy (Aydin, 2011).
Besides, social presence in e-learning could decrease the dropout rate (Cobb, 2009) and improve learners' e-learning performance such as critical thinking (Garrison, Anderson, & Archer, 2000) and online communications (Danchak, Walther, & Swan, 2001).E-learning performance has been demonstrated subject to several factors, e.g.motivation and learning strategies, computer competence, perceptions about discussion, critical thinking, peer learning, problem-based learning, interaction, and available help in a Chinese educational context (Zhu, Valcke, & Li, 2009).
Gender is, however, not considered a factor that in uences e-learning performance.There is no signi cant gender difference in language performance, while females show signi cantly higher self-e cacy than males (Harb, Bakar, & Krish, 2014).No gender difference has been found in e-learning via video modeling examples and both males and females experience an enhanced self-perceived competence after this elearning model (Hoogerheide, Loyens, & Van Gog, 2016).

Attitudes
Gender differences in attitudes toward e-learning are generally insigni cant although there are some different arguments.Students, whether males or females, hold positive attitudes towards the e-learning platform -e-HO in China (Lee, Pan, & Liao, 2011).Gender does not exert a signi cant in uence on attitudes towards e-learning (Chen & Tsai, 2007).Little evidence has been found regarding gender differences in attitudes towards e-learning system (Albert & Johnson, 2011).However, signi cant gender differences have been reported by some researchers (e.g.Jackson, Ervin, Gardner, & Schmitt, 2001;Shashaani & Khalili, 2001).Males tend to hold more positive attitudes (Whitely, 1997) toward e-learning and Chinese learners are more voluntary to access e-learning (Ong & Lai, 2006).Male university students prefer to use e-learning compared with females (Reda & Dennis, 1992).Males tend to hold more favorable attitudes towards e-learning than females and the latter hold more computer anxiety than the former (Keller et al., 2007) in Sweden and Lithuania.Females hold signi cantly more positive attitudes toward and are more interested in e-learning medical course with Moodle than males (Harreiter, Wiener, Plass, & Kautzky-Willer, 2011).
However, others found no gender differences in attitudes toward e-learning.They held that the super cial gender differences in attitudes may be caused by different social statuses, economic states and preferences rather than sex itself (e.g.Bimber, 2000) and gender differences in the attitude have been minimized with the rapid popularization of e-technologies and equally easy access to e-learning (Hanauer, Dibble, Fortin, & Col, 2004;Papastergiou & Solomonidou, 2005).For both genders, attitudes towards e-learning are positively correlated with their satisfaction in Cyprus, Thailand and other countries (Vate-U-Lan, 2020).No signi cant gender differences among university faculty and students have been found in attitudes toward information and communication technology assisted learning in a university in India (Verma & Dahiya, 2016).Chinese learners' attitude towards use of e-learning indicates the intention to use e-learning methods (Ong & Lai, 2006).No signi cant behavioral intention of e-learning has been identi ed between male and female instructors in Jordan (Altawallbeh, Thiam, Alshourah, & Fong, 2015).
Based on the review of literature, the research question proposed is "are there any gender differences in elearners' self-e cacy, satisfaction, motivation, attitude, and performance across the world?"

Methods
This meta-analysis is implemented on the basis of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Moher, Liberati, Tetzlaff, Altman, & The PRISMA Group, 2009).The review board waived the review protocol registry due to the characteristics of this study.

Eligibility criteria
The studies will be included if they (1) focus on gender differences in e-learning outcomes rather than elearning technology itself, (2) are of high quality based on University of West England Framework for Critically Appraising Research Articles (Moule et al., 2003), (3) adopt a randomized controlled design where a control and experiment group is comparatively analyzed, (4) can provide enough data for a metaanalysis, and (5) are written in English.
The studies will be excluded if they (1) focus on e-learning technology itself rather than e-learning outcomes, (2) are not rigidly designed using randomized controlled design, (3) are written in a language other than English, (4) cannot provide enough data for the meta-analysis even after corresponding with the authors, or (5) themselves belong to review studies.

Data sources and search strategy
To avoid duplication of this meta-analysis, we searched multiple databases e.g. the Cochrane Databases of Systematic Review, the Centre for Review and Dissemination, Taylor & Francis Group, Sage Publications, Springer Nature, Web of Science, Science Direct, EBSCO, and Educational Research Complete.To include as comprehensive literature as possible, we considered both published and unpublished literature written in English without time limitation.We included those ranging from their inception to February 10, 2021.We adopted a three-step search strategy.Firstly, we selected numerous databases such as Scopus, Taylor & Francis Group, Sage Publications, Springer Nature, Web of Science, Science Direct, Ebsco, Proquest, and Educational Research Complete.Secondly, we comprehensively searched the literature by entering corresponding terms into various databases and obtained results containing a sea of literature.Thirdly, we read through the literature to prevent duplication by optimizing the results.
The selection process of literature was implemented based on the PRISMA owchart (Figure 1).Firstly, the obtained results were entered into the software Endnote X8 (Thomson Reuters, New York, USA) for duplication identi cation and removal.Secondly, two reviewers screened the irrelevant literature by perusing abstracts, keywords, and titles, etc. Thirdly, both reviewers independently evaluated the literature for eligibility based on University of West England Framework for Critically Appraising Research Articles (Moule et al., 2003).Fourthly, both reviewers met together to decide the nal selection.In case both reviewers cannot reach an agreement on any selected literature, a third reviewer will join and determine the selection.

Quality assessment
The University of West England Framework for Critically Appraising Research Articles (Moule et al., 2003) evaluates each article in terms of ve sections, i.e.The Introduction, The Methods Section, Ethics, The Results/Findings, and the Conclusions.Each section is evaluated based on a given criterion.For example, as for the introduction part, reviewers will evaluate it by proposing criteria such as whether there is a clear statement about the topic being investigated and whether there is a clear rational for the research.As for the methods section, reviewers will evaluate it based on four criteria, i.e. (1) The research design should be clearly described; (2) The research methods should be appropriate for the topic being investigated; (3) The researchers should acknowledge the advantages or disadvantages of the design; (4) There should be a clear statement about how the participants were selected.Each article will be scored based on the criteria.Those top scored will be included for the meta-analysis.The results/ ndings section require that the results be related to the literature review and the researchers acknowledge the limitations of the research design.In the conclusion section, the researchers should acknowledge the implications for future research, identify areas for further research, and propose recommendations for practice from the results or discussions.

Data extraction
Both reviewers will extract speci c data from the included studies.The extracted data include total numbers of participants, means, and standard deviations in both control and experimental groups, levels of education of participants, modes of e-learning, countries where the study was conducted, e-learning outcomes (e-learners' attitudes, motivation, performance, satisfaction, and self-e cacy), and data collection methods.In case the data are not enough for the meta-analysis, we will correspond with the authors.The study will be removed if we nally fail to obtain enough data for the meta-analysis.The main extracted data are shown in Table 1.

Statistical analysis
We conducted the meta-analysis generally through Stata MP/14.0.Speci cally, we entered related data into Stata MP/14.0 to calculate standard mean differences (SMD) or Cohen d, the lower and upper bounds of 95% con dence intervals, weights, distribution of individual studies, Q data, heterogeneity, Isquared (I 2 ), p values and pooled results through forest plots.Cohen d is calculated as the mean difference between the experimental and control group divided by the standard deviation of the learning outcome across both groups (Sedgwick & Marston, 2013).
The statistics I 2 , calculated as the percentage of total variation of all included studies, was used to measure the heterogeneity of effect sizes.The heterogeneity is considered commonly existent in different studies.Thus, we measure it through Higgins & Green's criteria (2011), i.e. the heterogeneity will be considered unimportant if 0% < I 2 < 40%, moderate if 30% < I 2 < 60%, substantial if 50% < I 2 < 90%, and considerable if 75% < I 2 < 100%.If I 2 is larger than 50%, the results will prove signi cantly heterogeneous.
We will then adopt a random-effect model to conduct the meta-analysis.If I 2 is smaller than 50%, the results will prove insigni cantly heterogeneous.We will thus conduct the meta-analysis using a xedeffect model.Z statistics will be adopted to test the publication bias.The p value being smaller than .05indicates the presence of the publication bias while its being larger than .05indicates the absence of the publication bias.We also tested the publication bias via Begg's and Egger's tests through funnel plots where no-effect lines and individual studies are shown, as well as speci c effect sizes, and standard errors of effect sizes.
The symmetric distribution of dots along the no-effect line in a funnel plot indicates the absence of the publication bias while the asymmetric distribution indicates the presence of the publication bias.

Study selection
According to PRISMA flowchart (Moher et al., 2009), we obtained totally 12873 results from a number of databases, i.e.Taylor & Francis, Sage Publications, Springer Nature, Wiley, Elsevier, JSTOR, Web of Science, Science Direct, EBSCO, and Educational Research Complete.We obtained 1571 results after removing 11302 duplicated results via Endnote.Two reviewers selected 1189 results after independently screening and excluding 382 results after perusing abstracts, titles, and keywords.A total of 102 results passed the evaluation process.After removing 82 results due to various reasons such as incomplete data, improper design, and missing information, we selected 20 full texts.We then undertook the meta-analysis based on the included 20 studies, whose major characteristics are summarized in Table 1.

Characteristics of studies
As shown in Table 1, we summarize the main characteristics of included studies.The studies were conducted in various countries across the world, e.g.China, the USA, Austria, the Netherlands, Jordan, Chile, Spain, Malaysia, Indonesia, the UK, and India.The e-learning modes include a single e-learning course, multiple e-learning courses, inter-disciplinary e-learning courses, and various e-learning platforms.The educational levels of participants include university, elementary and secondary schools, and community college.The data collection methods include survey, pre-and post-tests, a written final assessment test, e-learning platforms such as e-HO, Moodle, and online English tests.The e-learning outcomes are classified into satisfaction, attitude, motivation, self-efficacy, and performance.The included studies can be classified into peer-reviewed journal articles, conference articles, and book chapters.

Tests of publication bias
To enhance the reliability of the results, we tested the publication bias using both Begg's and Egger's tests.As for Begg's test, we tested the publication bias using "metabias _ES _seES, begg" as a command to test the rank correlation between standardized intervention effect and its standard error (data input format theta se_theta assumed).The results indicate absence of publication bias (Kendall's Score (P-Q) = 144, Std.Dev. of Score = 227.36,z = 0.63, Pr > |z| = 0.529).
As for Egger's test, we entered the command "metabias _ES _seES, egger graph" into Stata MP/14.0 for detection of the publication bias since Egger's test can detect publication bias more sensitively than Begg's test (Egger et al., 1997).It is shown in Figure 2 that the studies are nearly symmetrically distributed along both sides of the regression line.We therefore conclude that the results indicate the absence of publication bias (t = -0.64,p = 0.523, 95% confidence interval = -3.49~1.79).

A sensitivity analysis
The sensitivity analysis is used to test the reliability or robustness of the meta-analysis via a leave-oneout method.If the leave-one-out method produces consistent results, then the meta-analysis will be considered robust or reliable.To conduct the sensitivity analysis, we entered "numbers of participants, means, and standard deviations" across both experimental and control groups for the metan-based influence analysis.As shown in Figure 3, the meta-analysis estimates are all positioned between the upper and lower bounds of the 95% confidence interval given a named study is omitted.We, therefore, conclude that the meta-analysis results are robust or reliable.

Gender differences in e-learners' self-efficacy in different countries
To determine whether a random-effect or fixed-effect model is used to run the meta-analysis of gender differences in e-learners' self-efficacy in different countries, we firstly tested the heterogeneity of the meta-analysis estimates via a forest plot through Stata/MP 14.0 (Figure 4).

Gender differences in e-learners' satisfaction in different countries
To summarize gender differences in e-learners' satisfaction in different countries, we drew a forest plot using Stata/MP 14.0 (Figure 5).

Gender differences in e-learners' motivation in different countries
To examine the pooled effect of gender differences in e-learners' motivation in different countries, we drew a forest plot using Stata/MP 14.0 (Figure 6).

Gender differences in e-learners' attitude in different countries
To examine gender differences in e-learners' attitude in different countries, we drew a forest plot using Stata MP 14.0 (Figure 7).

Discussion
Findings of this study are generally consistent with previous research.As for e-learners' self-e cacy, no signi cant gender differences have been revealed in all of the countries except the USA.Baylor & Kim's study (2004), conducted in the USA, concluded that females had signi cantly higher self-e cacy than males in the e-learning context.Female agents (around 61%) greatly outnumbered males (around 39%), which might have caused gender bias.The agents, merely representing gender-speci c features, might have led to results different from the real humans participants although agents did play an important role in e-learning experiments.Participants working with female agents might have been positively in uenced by their soft, encouraging voice and image, followed by enhanced self-e cacy.
We did not nd any signi cant gender difference in e-learners' satisfaction in different countries.Elearning, as an innovative learning method, has drawn many learners' attention whether they are biologically male or female.It could bring great convenience to them through the advanced information technologies.Learners do not need to carry any heavy learning materials with them and they can engage in learning wherever and whenever they want to.Through e-learning platforms, they can swiftly transfer a huge amount of data and easily have access to learning resources.They can also enhance their satisfaction with e-learning through frequent interactions with peers or teachers to solve di cult problems and arrange their learning activities.Teachers can gather enough data regarding students' feedback and decide teaching progress accordingly.This can improve both teachers' and students' satisfaction with the information technology assisted pedagogical approach.
No signi cant gender differences in motivation were revealed among e-learning participants.In the elearning environment, learners can manage their learning activities on their own.E-learning activities are no longer limited by the physical classroom and the face-to-face teacher.They can establish learning goals, select learning contents, and determine learning styles based on their own preferences.E-learning provides unprecedented learning resources and creates an innovative learning environment, where learners tend to be greatly motivated to join the learning activities since they can conveniently learn via various kinds of apps, texts, videos, audios, and technologies.The e-learning environment also bridges the gap of communication through online collaborations.Learners can seek help from peers and resort to teachers for enquiry of di cult questions at will.They can also determine the learning progress and styles based on their own preferences, rather than limited to a certain style or progress.In this way, their learning motivation is improved whether they are female or male.
In the USA, Jordan, and China, there are no signi cant gender differences in the attitudes towards elearning.Since both genders hold positive attitudes toward e-learning, designers and teachers may not need to cater the e-learning approach to a speci c gender but to other demographics such as economic status (Albert & Johnson, 2011).When designing the e-learning strategy, teachers can comprehensively consider age and experience of Internet use to popularize and improve the effectiveness of use of elearning approaches (Altawallbeh, Thiam, Alshourah, & Fong, 2015).Although no signi cant gender differences in attitudes were found toward e-learning, both genders hold lower levels of communication self-e cacy (Chu & Tsai, 2009).Communication skills, different from simple clicking, sur ng or glimpsing, may need complicated cognitive involvements such as coordination of nger and eye movements and mental processing (Chu, 2010).However, in Austria, India, and mixed countries (Chile and Spain), females hold signi cantly more positive attitudes towards e-learning than males.Females may join or initiate more communications with peers and teachers, hold more social presence, and thus feel more satis ed with e-learning activities, followed by more positive attitudes than males who tend to seek information rather than communication using the Internet (Gonzalez-Gomez, Guardiola, Martín Rodríguez, & Montero Alonso, 2012;Johnson, 2011).Males, mostly aiming at personal success and higher social status, tend to be isolated from their peers and involve into critical thinking although psychological researchers have proved no gender differences in their mental inborn feedback to surroundings (Salomone, 2007).The e-learning platform could provide learners with a large number of resources and opportunities, where females show signi cantly more intense interest in gender issues which tends to be criticized by males (Harreiter, Wiener, Plass, & Kautzky-Willer, 2011).Females may spend more time examining contents through the e-learning approach, leading to more positive attitudes than their male counterparts.
In general, females more positively evaluated e-learning than males since the pooled diamond is situated to the left of the no-effect line (Figure 7).Submerged in abundant information in the e-learning platform, females can be more interested in their favorite issues such as gender-related learning materials while males aim to seek information bene cial to their purpose.Females may concentrate more on the interesting issues than males who aim to seek information that can improve their social status.Concerning learning issues, females may show more interest than males since the former tends to aim at gender-based learning issues and acquire knowledge through communication and social presence while the latter aims at social rank issues (Harreiter, Wiener, Plass, & Kautzky-Willer, 2011).Males tend to be distracted by a sea of information in case they cannot nd the information they need.In the e-learning context, males are more likely to present personal information representing their social status, while females are more likely to enjoy the bene ts of social networking when social information is reduced.
Females pay more attention to learning and social process and less attention to members of a learning community than males (Flanagin, Tiyaamornwong, O'Connor, & Seibold, 2002).This may enhance female attitudes towards e-learning and reduce male positive evaluation of an e-learning method.
Signi cant gender differences in e-learning performance were found among students at the London School of Economics (the UK) and University of Valencia (Spain) (Cuadrado-García, Ruiz-Molina, & Montoro-Pons, 2010).Females signi cantly outperformed males.As the authors mentioned, females greatly outnumbered males, which may have caused bias of results.We failed to reveal any gender difference in e-learners' performance in other countries such as the USA, the Netherlands, Jordan, Malaysia, and China.The new decade has been witnessing dramatic development of information technologies.Both males and females nowadays have equally convenient access to e-learning approaches in most of the countries across the world.Both genders tend to perform similarly but in the elearning process, males pay more attention to the competitiveness in the course, while females regard the virtual classroom as an opportunity for online cooperative learning and cherish the cooperative e-learning environment (Arbaugh, 2002).Different preferences may have offset their different performance levels and cause insigni cant gender differences in e-learning performance.
The e-learning environment can greatly facilitate discussion and opinion sharing, which can promote e cient information exchange and cultivate social relations between males and females (Wang, Christina, & Zhao, 2007).Social constructivists (e.g.Derry, Gance, Gance & Schlager, 2000) argued that discussion and opinion sharing could help learners construct high-quality knowledge structures.Through an appropriate teaching design, teachers can encourage students to solve di cult problems and facilitate active debates by gathering them online.Through frequent interactions and intentional organization of the teacher, balanced numbers of males and females can form an effective learning community under the supervision and guidance of the teacher, where both males and females can mutually assist for knowledge acquisition.Discussion and opinion sharing can bridge the gap of communication between males and females.They can increase their knowledge and improve their social skills, conducive to favorable e-learning performance.Different characteristics of both genders may have offset the originally different performance levels through the interactive process in the e-learning process.

Conclusion Major ndings
This study, including 20 high-quality publications, meta-analytically examined gender differences in elearning outcomes, e.g.e-learners' self-e cacy, satisfaction, motivation, attitude, and performance across the world.Generally, there are no signi cant gender differences in e-learning learning outcomes.Exceptions are that females signi cantly outperformed males in Spain and the UK, that in Austria, India, and mixed countries (Chile and Spain), females hold signi cantly more positive attitudes towards elearning than males, and that in the USA, females present signi cantly higher self-e cacy than males.

Limitations
While study is rigidly designed and follows the PRISMA ow process, there are still a number of limitations.Firstly, this study merely includes publications written in English, which may have caused publication bias.Secondly, this study cannot reach all of the literature due to the limitation of the library resource.Thirdly, the included studies may have biases themselves, which may have caused bias of results.

Future research directions
Gender in e-learning is a hot issue in need of interdisciplinary research such as psychology, sociology, linguistics, education, and computation.Future research into the gender issue may adopt crossdisciplinary research methods except for a meta-analysis. Figures

Table 1 .
Characteristics of included studies