Effect of video styles on learner engagement in MOOCs

ABSTRACT Video lectures in massive open online courses (MOOCs) provide an opportunity to not only deliver instructional content but also engage learners. While there are many different styles of video lectures, it is not clear how video styles affect learner engagement. This study analysed and critiqued different typologies of video styles and classified MOOC video styles on a speaker-centric to media-centric spectrum. A total of 1372 survey responses were used for data analysis. The findings indicated that the ‘media-centric’ and ‘balanced’ video styles enhanced learner engagement to varying degrees in MOOCs of different study areas. In contrast, the ‘speaker-centric’ video style offered no advantages for promoting engagement in any MOOC study area. Effect sizes ranged from .03 to .07, indicating that video styles had a small to medium effect on engagement. These findings can provide new insights into the design of video lectures for different study areas in MOOCs.


Introduction
By the end of 2021, the number of universities that offer massive open online courses (MOOCs) had exceeded 950, the number of MOOC learners had exceeded 220 million, and the number of MOOCs had exceeded 19,000 (Shah, 2022).Early MOOC research tended to emphasise outcome measures valued in traditional higher education settings, particularly academic achievement (Alario-Hoyos et al., 2016;Kennedy et al., 2015) and retention (Greene et al., 2015;Pursel et al., 2016).The last three years have witnessed a shift from enhancing learning performance and retention to providing a more engaging learning environment for MOOC participants (Alemayehu & Chen, 2021;Romero-Rodríguez et al., 2019).The underlying rationale for this shift is that MOOCs are distinct from creditbearing online courses.MOOC learners display a variety of intrinsic, extrinsic, and social motivations (Deng et al., 2019).Topical interest is often the primary reason driving people to register for a MOOC (Ruipérez-Valiente et al., 2022).MOOC learners are more concerned with fulfilling personal agendas rather than performing well academically (Jung et al., 2019).Only a small percentage of learners intend to complete an entire MOOC and earn a certificate of completion (Semenova, 2022).More recent research has explored personal (e.g.Kuo et al., 2021;Sun et al., 2020) and environmental factors (e.g.Cobos & Ruiz-Garcia, 2021;Gallego-Romero et al., 2020) contributing to higher levels of learner engagement with MOOCs.
Video lectures are an integral part of MOOCs.MOOC videos play a critical role in assisting learners with understanding difficult concepts, making courses interesting, and responding to terms such as 'educational videos', 'instructional videos', 'lecture videos', 'MOOC videos,' and 'MOOC lectures', which are found in other research papers pertinent to multimedia learning.

Typologies of video styles
Over the last two decades, researchers have attempted to classify video styles into distinct categories (Table 1).The number of categories usually ranges from 2 to 18, and the labels given to categories in each taxonomy differ greatly.These differences can be attributed to the different classification rationales (e.g.learning outcomes, educational purposes) and the distinct research contexts in which a typology is developed (e.g.streaming, podcasting), as shown in Table 1.The purpose of analysing and critiquing these typologies is not to identify the best classification scheme but to select the most appropriate typology for use in this study.A review of recent literature reveals two main approaches to classifying video styles.One is classifying video styles in different or overlapping categories (e.g.Hansch et al., 2015); the other is defining video styles on one or more spectrums (e.g.Chorianopoulos, 2018).
Early typologies tended to divide video styles into different or overlapping categories.These typologies provide useful heuristics for distinguishing mainstream video styles.A closer inspection of these classification schemes, however, shows that many categories in fact carry the same meaning.For instance, 'presentation' (Majid et al., 2012), 'enhanced video podcast' (Kay, 2012), 'slides' (Guo et al., 2014), and 'presentation slides with voice-over' (Hansch et al., 2015) all refer to the same style, featuring slide shows coupled with audio explanations.Similarly, 'lecture-based video podcast' (Kay, 2012), 'lecture' (Majid et al., 2012), 'classroom' (Guo et al., 2014), and 'classroom lecture' (Hansch et al., 2015) are used to describe educational videos captured in live classroom settings.The use of different labels to describe similar video styles may cause confusion and is not conducive to investigating the effectiveness of video styles.
Apart from the proliferation of terms, there are also other potential issues associated with existing classification schemes.Categories within some typologies have a narrow scope.For example, 'Khanstyle tablet capture' and 'Udacity-style tablet capture' in the typology of Hansch et al. (2015) are both concerned with instructors drawing content (e.g.mathematical formulas, short texts) on a virtual whiteboard.Classification schemes with a narrow scope run the risk of increasing repetition and may have poor predictive attributes for defining new video styles that do not yet exist.Additionally, categories in some typologies are neither exhaustive nor mutually exclusive.The criteria for creating these typologies appear to be ad hoc and are not thoroughly explained.For instance, Majid et al. (2012) and Guo et al. (2014) did not explain how they classified video styles and ignored common styles such as animation.More recent typologies arrange video styles on spectrums as opposed to classifying them into distinct categories.Santos Espino et al. (2016) placed MOOC video styles on a speaker-centric to board-centric spectrum.Santos Espino et al. (2016) defined board-centric as videos characterised by a rectangle-shaped surface where instructional content is presented.In board-centric videos, the board fills the full frame or a large frame area, and the visual information (e.g.texts, graphics) is primary.Meanwhile, Santos Espino et al. (2016) defined speaker-centric as videos featuring one or more visible human speakers as the main vehicle for delivering instructional content.In speakercentric videos, the speakers are visible most of the time and tend to provide oral information.According to Santos Espino et al. (2016), board-centric styles include virtual whiteboards, screencasts, and slides, while examples of speaker-centric styles include talking head videos, interviews, and live lecture recordings.The speaker-centric to board-centric spectrum simplifies category-based classification schemes and has the potential to accommodate novel video styles.
Building on Santos Espino et al. (2016), Chorianopoulos (2018) also placed educational video styles on spectrums.Noting that Santos Espino et al. (2016) presented video styles as opposing conditions in only one dimension, Chorianopoulos (2018) argued that speakers and boards do not compete for the learner's attention.Based on this assumption, Chorianopoulos (2018) theorised that the styles of educational videos are determined by two complementary dimensions -human embodiment and instructional media -and there are multiple values that range from the digital to the physical spectrum for each of the two dimensions.For example, one can find the video style characterised by presentation slides with a talking head at the mid-range of the instructional media spectrum and at the physical end of the human embodiment spectrum (Figure 1).The strength of Chorianopoulos's (2018) work is that the taxonomy defines a two-dimensional space of existing and potential new video styles.The classification scheme considers not only the presence of humans and instructional media but also the level of human presence and the type of instructional media.
The classification schemes of Santos Espino et al. (2016) and Chorianopoulos (2018) are both potentially useful for comparing the styles of educational videos.Santos Espino et al. (2016) built their taxonomy based on an empirical examination of frequently used video styles in MOOCs.Chorianopoulos's taxonomy (Chorianopoulos, 2018) is future-proof and is more useful for identifying new video styles that do not yet exist.A key characteristic of a typology is that its dimensions can be based on the notion of an ideal type and are not necessarily something found in empirical reality (Smith, 2002).Until now, very few MOOCs have adopted virtual agents (e.g.animated characters, robots) for video instruction.Chorianopoulos (2018) also noted that almost no videos found in major educational video repositories employed artificial agents to manipulate instructional media.The present study aimed to compare video styles frequently adopted in MOOCs.For the above reasons, the taxonomy of Santos Espino et al. (2016) was selected for this study.This study renames 'boardcentric' as 'media-centric' to de-emphasise the rectangular shape and highlight the nature of the instructional content displayed on the board.A 'balanced' midpoint was added to represent video styles with a combination of degrees of speaker-and media-centricity.A speaker-centric to mediacentric spectrum is used throughout the study (Figure 2).

Video styles and learner engagement
This study conceptualised learner engagement as a multidimensional construct, comprising behavioural, cognitive, emotional, and social engagement.Guo et al. (2014, p. 43) argued that 'true engagement is impossible to measure without direct observation and questioning' and adopted two proxies for engagement: engagement time and assignment submission.This type of engagement is often referred to as behavioural engagement, which is conceptualised as learners' observable actions and their participation and involvement in educational activities (Appleton et al., 2006).Behavioural engagement in MOOCs is typically measured through video use, user progression, and completion of assessment tasks (Deng et al., 2019) and has been found to be predictive of desirable learning outcomes such as completion (Jung & Lee, 2018).Studies of learner engagement in MOOCs have tended to focus on behavioural engagement because behaviour is overt and relatively easy to identify and track (Deng et al., 2019).Video lectures are potentially useful for enhancing behavioural engagement in MOOCs.Although Costley and Lange (2017) noted that the style of instructional videos was not correlated with the number of videos viewed by university students, Guo et al. (2014) found that MOOC videos that alternated between a speaker's talking head and slides prompted learners to spend more time on watching video lectures than MOOC videos only presenting PowerPoint slides and code screencasts.Unlike university courses that have face-to-face teaching/contact time and utilise videos as supplementary materials (Manuel et al., 2021), MOOCs do not usually organise face-to-face sessions and are mostly built around a series of instructional videos (Deng et al., 2019;Lemay & Doleck, 2022).In the MOOC context, it is reasonable to speculate that the 'balanced' video style featuring both the speaker and media may be more effective in promoting behavioural engagement than the 'mediacentric' and 'speaker-centric' video styles, for it provides some social and emotional clues to help learners make sense of instructional content presented in an almost entirely virtual learning environment (Walji et al., 2016).The current study hypothesises the following: The level of behavioural engagement is higher in MOOCs adopting the 'balanced' video style than the MOOCs adopting the 'media-centric ' and 'speaker-centric' video styles (H1).
Social engagement, representing learner-instructor and learner-learner interactions in MOOCs (Tawfik et al., 2017), is sometimes confounded with behavioural engagement.This is because interacting with instructors and other learners is observable and part of the 'way of behaving' in educational activities (Maroco et al., 2016).Recent research has shown that social engagement (e.g.learner-learner interaction) is a separate construct from behavioural engagement (e.g.learnercontent interaction) in MOOCs (Deng et al., 2020).Social engagement is associated with desirable MOOC outcomes such as academic performance and course completion (Deng et al., 2019).However, it remains unclear from the extant literature whether certain video styles lead to improved social engagement.This study aimed to fill this knowledge gap and investigate the effects of video styles on social engagement.
Video lectures could be a potentially useful instructional strategy for improving social engagement in MOOCs.MOOC participants often post comments and questions directly related to a specific section of a video lecture (Chen et al., 2019).Individuals tend to experience a higher quality of social interaction in online learning environments when different types of media are employed and when the various forms of media complement each other (Kim et al., 2011).It is therefore reasonable to suspect that the 'media-centric' video style may be more effective in sustaining social engagement in MOOCs than the 'balanced' and 'speaker-centric' video styles.Because of limited usage of media formats, the 'speaker-centric' video style may not play a facilitative role in improving social engagement.This study thus formulates the following hypothesis: The level of social engagement is higher in MOOCs adopting the 'media-centric' video style than the MOOCs adopting the 'balanced ' and 'speaker-centric' video styles (H2).
Some researchers have drawn scholarly attention to engagement domains that are less observable and more internal.Li and Baker (2018), for example, revealed that a large proportion of MOOC participants who watched video lectures were not actively processing the instructional content in the videos.In other words, being behaviourally engaged in MOOCs does not necessarily mean being cognitively engaged.Cognitive engagement refers to individuals' willingness to exert mental effort to comprehend complex ideas, master difficult skills, and strengthen learning and performance (Appleton et al., 2006).Cognitive engagement in MOOCs is linked to positive outcomes, such as academic performance (Li & Baker, 2018) and perceived learning (Lan & Hew, 2020).The present study conceptualised cognitive engagement as an essential domain and explored the effects of video styles on cognitive engagement.
The 'balanced' video style may be effective in promoting MOOC learners' cognitive engagement.Eye-tracking research indicated that the presence of the speaker resulted in a higher percentage of fixation on the speaker and improved learners' academic performance, suggesting that the processing of the speaker's image may have provided social cues to facilitate the processing of cognitively relevant information (Wang et al., 2020).The speaker's guided gaze, for example, contributed to a longer dwell time on the content area and enhanced learning performance (Pi et al., 2020).This is likely because the speaker successfully drew learners' attention to important information in the learning materials (Deng & Gao, 2023b).However, the level of effectiveness may be dependent on the fields of study.A study indicated that presenting the speaker in video lectures increased individuals' cognitive load when learning procedural knowledge, a type of knowledge about how to perform a task, but not when learning declarative knowledge, the knowledge of facts and rules (Hong et al., 2018).It is reasonable to infer that the 'balanced' video style may be more effective in promoting cognitive engagement than the 'media-centric' video style, but the degree of effectiveness can vary across study areas.The 'speaker-centric' video style may not facilitate cognitive engagement well, for it presents important information through verbal communication only.The following hypothesis is formulated: The level of cognitive engagement is higher in MOOCs adopting the 'balanced' video style than the MOOCs adopting the 'media-centric ' and 'speaker-centric' video styles (H3).
Another less overt engagement domain is emotional engagement.Emotional engagement is centred on the emotional connections students make with instructors, peers, and content (Appleton et al., 2006).MOOCs afford the creation and distribution of various emotions.Textual emotional expressions, such as altruistic emotions and negativity (Comer et al., 2015), were observed in MOOC discussion boards.MOOC video lectures also induce positive (e.g.excited, interesting) and negative (e.g.confused, disappointed) emotions (Chen et al., 2017).Emotional engagement in MOOCs was found to positively affect perceived learning (Lan & Hew, 2020).Video lectures provide an important avenue for eliciting positive emotions from MOOC participants (Deng & Benckendorff, 2021;Deng & Gao, 2023c); however, it is not known whether video styles quantitatively affect MOOC participants' emotional engagement.The current study aimed to fill this knowledge gap by investigating the effects of video styles on emotional engagement.
The 'balanced' video style may be effective in strengthening emotional engagement with MOOCs.A recent meta-analysis showed that video lectures incorporating the speaker's face had a positive, significant effect on motivation to study (Alemdag, 2022).Eye-tracking research shows that the speaker's happy face did not contribute to a longer dwell time on the content or the speaker area, but learners' academic achievement still improved (Pi et al., 2022).This is probably because the speaker elicited beneficial social-emotional responses from learners.It is reasonable to speculate that the 'balanced' video style may be more effective than the 'media-centric' one in facilitating motivational or affective processes during the study of MOOCs.Emotional processes are inseparable from cognitive processes in multimedia learning (Deng & Gao, 2023b;Plass & Kaplan, 2016).The 'speakercentric' video style may not be as effective in promoting emotional engagement.This is likely because a lack of textual or visual representations of knowledge on screen may hinder the information selection, organisation, and integration processes required for effective multimedia learning, which in turn may hamper motivational and affective processes (Endres et al., 2020).The following hypothesis is proposed: The level of emotional engagement is higher in MOOCs adopting the 'balanced' video style than the MOOCs adopting the 'media-centric' and 'speaker-centric' video styles (H4).

Data collection
Prior to data collection, ethics clearance was obtained from the human research ethics committee of the researcher's university.A self-administered survey was used for data collection.Respondents were asked to nominate a MOOC they had recently participated in and keep that MOOC in mind when completing the survey.A screening question was used to filter out respondents who reported that videos were not used in their nominated MOOC.To minimise the memory effect, individuals who participated in a MOOC in the previous 12 months were selected for data analysis.If TECHNOLOGY, PEDAGOGY AND EDUCATION a respondent had participated in a MOOC in the last 12 months, they would be directed to the instruction page to fill out the survey.The instruction asked the respondents to think about the most recent MOOC they studied and keep this MOOC in mind when answering all the questions in the survey.The researcher conducted a pilot study with 20 volunteers prior to formal data collection.Results of the pilot study showed that the instructions were clear, and the survey items were comprehensible.Two questions were tweaked based on feedback from the volunteers to make the survey easier to answer.
The convenience sampling approach was adopted for two reasons.First, this research is one of the first scholarly endeavours to empirically investigate the relationship between video styles and learner engagement in MOOCs, and it is exploratory in nature.Second, the population and population composition of MOOC users are not known, and MOOC platforms are usually unwilling to disclose the personal characteristics of their users for privacy reasons, making it difficult to employ probability sampling methods to systematically access MOOC learners.
Participants in this study were individuals who studied one or more MOOCs offered by a researchintensive university on edX and were recruited through email newsletters.This university is an edX charter member and has contributed more than 60 courses to the edX platform.The newsletter contained a brief introduction to the study and the survey URL.Participants were also recruited from social media websites to incorporate learners across a diverse range of MOOCs and MOOC platforms.Data collection lasted a total of four weeks.All the respondents signed the informed consent before responding to the survey.A total of 1372 complete cases were used for data analysis after deleting outliers, surveys with incomplete responses, and unqualified respondents.
All the 1372 respondents had participated in at least one MOOC in the previous 12 months and confirmed the MOOCs they studied used instructional videos as the medium of instruction.The participants had relatively diverse ethnic origins -over 35% of the individuals came from Asia, followed by Africa (15.0%),North America (13.8%),Europe (13.3%),Latin America and the Caribbean (12.2%), the Arab states (5.6%), and Oceania (4.8%).Of the respondents, 63.5% were male.The predominant age groups were 25-34 years old (35.0%), followed by under 25 (34.3%),35-44 (16.0%), 45-54 (7.6%), and 55 and over (7.1%).Those with at least a bachelor's degree comprised a large cohort (72.7%), which is consistent with Deng et al's. (2019) observation that MOOC participants tended to be qualified, experienced learners who already hold a bachelor's or master's degree.

Measures
Video styles were treated as the predictor variable and measured at the nominal level.The classification of video styles in this study was built based on Santos Espino et al. (2016).The researcher arranged MOOC video styles on a speaker-centric to media-centric spectrum and developed a question to obtain information about the main style of video lectures used in the respondent's nominated MOOC.The predictor variable consisted of three categorical, independent groups: 'media-centric', 'speaker-centric,' and 'a balance between media-centric and speaker-centric'.The researcher explained the three styles in greater detail to help respondents fully understand the options in the survey.For example, 'speaker-centric' indicates that a human speaker is visible most of the time, with little or no multimedia content.To further help respondents identify the main style of the videos utilised in the MOOC they participated in, four illustrations were provided to accompany the text description for each video type.The illustrations were taken from MOOC video lectures in various disciplines on Coursera.The researcher acknowledges that the present study measured perceptions of video styles.The interpretation of the findings is therefore based on the premise that it is learners' perceptions of video styles that affect engagement, and actions should be taken to design video lectures that are congruent with learners' perceptions of video styles.The observations between groups were independent, meaning that each group was made up of different MOOC participants.
Learner engagement was treated as the outcome variable and measured at the ordinal level.Learning analytics were not used as proxies for engagement in this study for two reasons.On the one hand, learning analytics only capture individuals' engagement with online materials (Baker et al., 2020).Researchers are likely to miss important insights if learners use alternative study strategies, such as downloading video lectures from the learning management system.On the other hand, learning analytics often oversimplify the learning process by equating observable behaviours with learner engagement (Guzmán-Valenzuela et al., 2021).Past research has shown that the extrapolation of cognitive and emotional engagement from learning analytics can be both laborious and inaccurate (Li & Baker, 2018).
In view of these challenges, this research employed the MOOC Engagement Scale (MES) to capture learner engagement in MOOCs.The MES was developed and validated by Deng et al. (2020), and it showed desirable reliability, face validity, content validity, construct validity, convergent validity, and discriminant validity.This instrument has been recently applied and/or adapted in a number of educational studies (e.g.Deng & Gao, 2023a;Jiang & Peng, 2023), and it demonstrated good validity and reliability (e.g.Hoi & Le Hang, 2021;Xu et al., 2023).
The MES contained 12 statements.A 6-point Likert scale ranging from strongly disagree (1) to strongly agree (6) was used.Each engagement domain was surveyed using three statements.In this study, the Cronbach's alpha values for the behavioural, cognitive, emotional, and social engagement items were .73,.71,.76,and .83,respectively, demonstrating good internal consistency.To facilitate the discussion of the key findings, the operational definition for each engagement domain is provided below.
• Behavioural engagement: participants study a MOOC regularly, take notes while studying the MOOC, and review notes when preparing for assessments.• Cognitive engagement: participants' mental investment in the study of a MOOC to comprehend complex ideas, master difficult skills, and strengthen learning and performance.• Emotional engagement: the degree to which participants are inspired to expand knowledge in a MOOC and enjoy studying the MOOC.• Social engagement: learning-related interactions between participants and instructors or other learners, on and beyond the MOOC platform.
The effects of video styles on learning and learner perceptions can vary across fields of study.This is because each study area has its preferred way of disseminating information and knowledge (Santos Espino et al., 2016Espino et al., , 2020)).In view of these differences, the researcher controlled for the study area when investigating the relationship between video styles and learner engagement.The researcher divided study areas into five categories: 'arts and humanities', 'computers, engineering, and mathematics', 'natural sciences', 'social sciences,' and 'personal development'.The categorisation process was based on course catalogues on Coursera and FutureLearn.To help respondents choose the most relevant study area, the researcher provided examples for each option.For instance, respondents were informed that the 'arts and humanities' study area incorporates subjects such as design, history, languages, literature, music, theology, religion, and philosophy.An open-ended question was designed so that respondents who had difficulty identifying the study area could manually input the data.
The researcher then reclassified the manually entered data.No new category emerged during the reclassification process.
The researcher used four questions to collect demographic information from the respondents.Previous research has revealed that age, education background, gender, and origin were the most frequently reported demographic variables in the MOOC literature (Deng et al., 2019).These factors were compared with recent MOOC publications to ensure that the current study represents a relatively diverse sample.

Data analysis
Descriptive statistics on demographics were calculated using IBM SPSS 26.Principal component analysis (PCA) was implemented to single out items that could be conceived as indicators of behavioural, cognitive, emotional, and social engagement.To comply with the principle of parsimony and reduce measurement error, the researcher created four aggregated rating scales by calculating the sum of the variables comprising each engagement domain of the MES.The original 12 engagement items were substituted by the four aggregated rating scales, ranging from 0 to 18. Aggregated rating scales not only overcome the measurement error inherent in the measured variables to a certain extent, but also represented the multiple facets of a concept in a single measure.The composite scores derived from PCA were retained for further use in the Kruskal-Wallis tests and post-hoc analysis.
Next, Kolmogorov-Smirnov and Shapiro-Wilk tests were performed to check if values on the aggregated scales fit a normal distribution.Results showed that values on the aggregated scales were non-normal.Therefore, non-parametric tests were applied.A series of Kruskal-Wallis tests were undertaken to determine the relationship between video styles and learner engagement in MOOCs.Mann-Whitney U tests were employed for post-hoc pairwise comparisons when the results of the Kruskal-Wallis test showed a significant difference at the .05level.To reduce the probability of type I errors, the results of the Mann-Whitney U tests were Bonferroni corrected.An inspection of histograms indicated that the sample distributions were similar in shape, but the shape was asymmetric.In view of this, the median is reported in the results section, and significant p-values were interpreted as rejecting the assumption of medians.

Results
Table 2 shows the demographic information of the MOOC participants (i.e.age, education background, gender, and origin).This research compared the sample composition with that of other MOOC studies to ensure certain subgroups of MOOC learners were not underrepresented or overrepresented.The participants in this study represented a relatively diverse sample and had demographic information comparable to that reported in other recent MOOC studies (e.g.Jung et al., 2019;Li, 2019).A total of 1372 cases were used for PCA.The number of observations per variable in this research met the desired ratio of 5:1 recommended by Hair et al. (2014).To determine the items making of each engagement domain, the researcher conducted PCA with varimax orthogonal rotation.Varimax orthogonal rotation was selected because the goal was to reduce data to a small number of variables for subsequent use in Kruskal-Wallis tests.This rotation method loaded a relatively smaller number of variables onto each factor, resulting in more interpretable clusters of factors.The best solution was a four-factor model accounting for 68.52% of the common variance (Table 3).The model contains a representative and parsimonious set of factors.The communalities and factor loadings were all > .40,which are considered to meet the requirement for interpretation of the structure (Hair et al., 2014).No factors were eliminated in the analysis process owing to low factor loadings or unacceptable communality values.The Kaiser-Meyer-Olkin measure of sampling adequacy index of .84 and a significant chi-square value for Bartlett's test of sphericity, χ 2 (66) = 5712.55,p < .001,suggested that the factor model was appropriate for the data.The factorial dimensions in this study were consistent with Deng et al's.(2020) instrumentation of learner engagement in MOOCs.The Cronbach's alpha values ranged from .71 to .83,exceeding a threshold of .70 (Hair et al., 2014).The result demonstrated a high degree of reliability of the scales.The composite scores were calculated based on the results of PCA and were treated as outcome variables in the Kruskal-Wallis tests.
The researcher performed Kruskal-Wallis tests to investigate whether video styles affected learner engagement in arts and humanities MOOCs.This procedure was followed by applying post-hoc Mann-Whitney tests to compare differences in the level of engagement between two independent groups.Table 4 shows that the relationship between video styles and social engagement was statistically significant, H(2) = 6.50, p = .039.Post-hoc Mann-Whitney tests using a Bonferronicorrected alpha level of .017were adopted to compare all pairs of groups.The difference in social engagement between the 'media-centric' (Mdn = 10.5) and 'balanced' groups (Mdn = 8.0) was statistically significant, U = 2247.50,Z = −2.60,p = .009.None of the other comparisons were significant after Bonferroni adjustment.
Kruskal-Wallis tests were also performed to investigate whether video styles affected levels of engagement in social sciences MOOCs (Table 7) and personal development MOOCs (Table 8).Results showed that statistically significant differences were absent for the four engagement domains, indicating that video styles were independent of learner engagement in social sciences and personal development MOOCs.
The Kruskal-Wallis tests revealed statistically significant differences in engagement levels for the three groups of MOOC participants who reported the use of different video styles ('media-centric', 'speaker-centric', 'balanced') (Table 9).The differences were observed in all study areas, except social science and personal development MOOCs.From the Bonferroni-corrected Mann-Whitney post-hoc tests, this study found that video styles were related to cognitive, emotional, and social engagement but not behavioural engagement.Specifically, the 'balanced' group exhibited higher cognitive and emotional engagement than the 'speaker-centric' group in computers, engineering, and mathematics MOOCs.The 'media-centric' group showed higher social engagement than the 'balanced' group in arts and humanities MOOCs and computers, engineering, and mathematics MOOCs.The 'media-centric' group also exhibited higher social engagement than the 'speaker-centric' group in natural sciences MOOCs.Eta-squared (η 2 ) indicates the percentage of variance in the ranks explained by the predictor variable, namely video styles.The η 2 ranged from .03 to .07,indicating a small to Note: NS: not significant.
TECHNOLOGY, PEDAGOGY AND EDUCATION medium effect size (Cohen, 1988).The summary of the hypotheses testing results is shown in Table 10.The results indicate that H1 is not supported, and H2, H3 and H4 are partially supported.

Discussion
Prior to this study, few attempts had been made to explore how video styles affect learner engagement in MOOCs.Most notably, Guo et al. (2014) found that MOOC videos characterised by a balance between slides and the talking head of the instructor contributed to higher levels of behavioural engagement.The results of this study did not support Hypothesis 1, which predicted that the 'balanced' video style would be more effective in promoting behavioural engagement in MOOCs.The present study found that video styles had no significant effects on behavioural engagementa finding that appears to contradict Guo et al's. (2014) research.A closer examination of the two studies, however, reveals that the inconsistency in the findings could stem from the different approaches to defining engagement.Guo et al. (2014) measured behavioural engagement through two proxies: engagement time and assignment submission.The present study used three indicators to represent behavioural engagement: studying a MOOC regularly, taking notes when studying MOOCs, and reviewing notes when preparing MOOC assessments.
The different relationships between video styles and behavioural engagement highlight the importance of clearly defining the scope of engagement upfront in academic research.The most commonly adopted indicators of behavioural engagement in MOOCs are the number of video lectures watched and assignments submitted by learners (Li & Baker, 2018).However, recent research has shown that simply counting learning activities is not an accurate indication of behavioural engagement since it does not reflect the quality of learning (Deng et al., 2020;Lee, 2018).For this reason, the current study de-emphasises the number of videos watched and the number of assignments attempted when evaluating behavioural engagement.Many scholars have noted the Hypotheses Results H1: The level of behavioural engagement is higher in MOOCs adopting the 'balanced' video style than the MOOCs adopting the 'media-centric' and 'speaker-centric' video styles.
Not supported H2: The level of social engagement is higher in MOOCs adopting the 'media-centric' video style than the MOOCs adopting the 'balanced' and 'speaker-centric' video styles.
Partially supported H3: The level of cognitive engagement is higher in MOOCs adopting the 'balanced' video style than the MOOCs adopting the 'media-centric' and 'speaker-centric' video styles.
Partially supported H4: The level of emotional engagement is higher in MOOCs adopting the 'balanced' video style than the MOOCs adopting the 'media-centric' and 'speaker-centric' video styles.
Partially supported need for greater engagement in MOOCs but have failed to provide a conceptual or operational definition of engagement (e.g.Sanz-Martínez et al., 2019).A clear definition of engagement will add conceptual clarity and facilitate the discussion of this topic across evidence-based studies.MOOC researchers have tended to emphasise the behavioural aspect of learner engagement (Wei et al., 2021).The present study maintains that MOOC participants manifest their engagement in different ways, and it extends the scope of engagement to incorporate behavioural, cognitive, emotional, and social domains.This research provided evidence that the 'media-centric' video style was associated with higher levels of social engagement compared to 'balanced ' and 'speaker-centric' video styles, but this association varied across study areas.The level of social engagement was higher in arts and humanities MOOCs, computers, science, and mathematics MOOCs, and natural sciences MOOCs.These results partially support Hypothesis 2. Social engagement can alleviate feelings of being isolated in the MOOC learning space and help reduce the dropout ratio (Wang et al., 2019).In arts and humanities MOOCs, for example, it is possible that the 'media-centric' video style characterising images, cultural artefacts, and historical materials provided pedagogically relevant information that induced and facilitated social engagement.If the aim is to promote social engagement, practitioners are advised to consider the 'media-centric' style as a priority in the production of MOOC video lectures.Recently, Hughes et al. (2019) and Lange and Costley (2020) found that exposing students to multiple types of media during video instruction was conducive to learning.A promising line of research would be to investigate how to combine a range of media types in MOOC video lectures to bolster social engagement.Some MOOC participants are not motivated to interact with other people and intend to learn from videos only (Daniels et al., 2016).Future research could explore whether video lectures designed for enhancing social engagement show effectiveness in this learner cohort.
This study also revealed that the 'balanced' video style was linked to higher levels of cognitive and emotional engagement compared to the 'speaker-centric' video style.However, this relationship only held true for the study area of computers, engineering, and mathematics.These results partially confirm Hypotheses 3 and 4. Cognitive engagement and emotional engagement are relatively under-researched topics in the MOOC literature (Deng et al., 2019).Recent research has shown that cognitive and emotional engagement had a positive influence on perceived learning (Lan & Hew, 2020).Cognitive engagement was also found to predict academic performance, particularly for individuals who followed the learning pathway intended by the MOOC instructor (Li & Baker, 2018).If practitioners aim to enhance cognitive and emotional engagement in computers, engineering, and mathematics MOOCs, the 'balanced' video style could be more efficacious and is preferred over the 'speaker-centric' video style.The different relationships between video styles and engagement across academic fields also explain why video styles have affected learning and learner perceptions in certain studies (e.g.Wang et al., 2020) but not in others (e.g.van Wermeskerken et al., 2018).In view of this, future research should consider disciplinary differences when exploring the effects of video styles on learning processes and outcomes.
This study found that the 'speaker-centric' video style had no advantages over 'balanced' or 'media-centric' video styles for enhancing learner engagement in any MOOC study area.A possible explanation is that 'speaker-centric' videos provide instructional content through verbal communication only, whereas 'balanced' and 'media-centric' videos provide content through both verbal and visual channels.The cognitive theory of multimedia learning proposes that students are able to build two mental representations (i.e.verbal and visual) and make connections between the two, and it is better to present an explanation using two modes of representation rather than one (Mayer & Moreno, 2002).In the higher education literature, several studies have compared the instructorpresent condition with the instructor-absent condition.These studies tend to agree that (a) instructor-present videos attract a considerable amount of visual attention, and (b) university students prefer and are more satisfied with instructor-present videos (Wang et al., 2020;Wilson et al., 2018).The present study shows that instructor-present videos can be further split into 'speaker-centric' and 'balanced' video styles, and the 'speaker-centric' video style should be used with discretion when prompting engagement is an important consideration.Despite this, it is possible that the relationship between video styles and engagement is moderated by learner characteristics.Recent research has shown that personal characteristics could be important boundary conditions to determine how video design principles work with different types of learners (Deng & Gao, 2023b).Individuals with different cultural backgrounds, for example, may exhibit different styles for processing visual stimuli (Haensel et al., 2022).In future work, investigations of differences in personal characteristics, such as country of origin, would provide additional insight into the effect of video styles on learner engagement.

Limitations and future direction
This study has some potential limitations along with associated considerations for future research.Firstly, as a preliminary study, this research employed a non-probability sampling approach.Although the participants in this research represented a relatively diverse sample and were comparable to the sample compositions reported in recent MOOC studies, the relationship between video styles and learner engagement is effectively exploratory and may not be generalisable to the entire MOOC population.Future research could aim to collaborate with MOOC platforms so that the population data could be obtained.This would allow for the use of stratified random sampling or random cluster sampling and improve the generalisability of the findings.To provide insight into the generalisability of the findings, future research should also consider potential interactions among video styles, learner engagement, and personal characteristics such as cross-cultural differences in processing audio-visual information.
Secondly, this study asked learners to report the main style of video lecture in their nominated MOOC.Although the researcher provided a detailed description and multiple illustrations to explain each option, there could have been slight variations in respondents' perceptions of each type of video style.Future research could overcome this potential limitation by employing additional mechanisms to further validate respondents' identification of the video style used in the MOOCs they participated in.Alternatively, future research could manipulate video styles as a predictor variable in a controlled laboratory study.
Finally, this study divided MOOC study areas into five categories.It should be acknowledged that there are other useful approaches to classifying fields of study.The researcher recommends considering nuanced disciplinary differences when investigating this topic in future research.

Conclusions
This study explored the effects of video styles on learner engagement in MOOCs.The objective was achieved by classifying MOOC video styles on a speaker-centric to media-centric spectrum, conceptualising and operationalising learner engagement as a multidimensional construct comprising four discrete domains, and empirically investigating the effects of video styles on learner engagement after controlling for the MOOC study area.The findings revealed that video styles affected MOOC participants' cognitive, emotional, and social engagement but not behavioural engagement.Specifically, the 'media-centric' video style was linked to higher levels of social engagement in all study areas, except social sciences and personal development.The 'balanced' video style was associated with higher levels of cognitive and emotional engagement, but this relationship only held true for computers, engineering, and mathematics MOOCs.The 'speaker-centric' video style offered no advantages over the 'balanced' or 'media-centric' video styles.The effect sizes ranged from .03 to .07,indicating that video styles had a small to medium effect on learner engagement.
This study makes several theoretical contributions to the literature.First, this study critiqued the existing typologies of video styles and classified MOOC video styles on a speaker-centric to mediacentric spectrum.Not only does this spectrum have conceptual, heuristic value, but it was also empirically employed to investigate the effects of video styles on learner engagement in MOOCs.Second, this study defined and operationalised learner engagement as a multidimensional construct comprising four discrete but related domains.Prior research tended to simplify learner engagement in MOOCs as overt, observable behaviours and ignore engagement domains that are more internal and less observable (e.g.Perna et al., 2014), or adopt a unified approach and measure engagement as the sum of several engagement domains (e.g.Chen & Thomas, 2020;Jung & Lee, 2018).This study confirmed the four-factor structure of learner engagement in MOOCs and provided a more nuanced understanding of the role of each engagement domain.Last, the present study investigated the effects of video styles on learner engagement with MOOCs and controlled for the MOOC study area when exploring such a relationship.The study differentiates MOOCs and credit-bearing university courses and shifts the focus from traditional outcome measures such as achievement and retention to learner engagement.The study results showed that the relationship between video styles and engagement was different across academic fields, highlighting the importance of considering disciplinary differences when evaluating the design of educational multimedia resources (Deng & Gao, 2023b).
Selecting the right video style is conducive to providing a more engaging learning environment for MOOC learners.The results of this study have implications for practitioners such as instructional designers and video producers.First, this study showed that the 'media-centric' video style, characterised by instructional content (e.g.text, pictures, animation) with little or no visible speaker, was linked to higher levels of social engagement.Practitioners are advised to prioritise the 'media-centric' style in the production of video lectures for (a) arts and humanities MOOCs; (b) computers, science, and mathematics MOOCs; and (c) natural sciences MOOCs.Second, this study revealed that the video style featuring a balance between instructional content and human speakers was associated with higher levels of cognitive and emotional engagement; however, this relationship only held true for the study area of computers, science, and mathematics.That is, the 'balanced' video style is preferred if the aim is to enhance cognitive and emotional engagement in computers, science, and mathematics MOOCs.Third, this study found that the 'speaker-centric' video style offered no advantages over the 'media-centric' or 'balanced' video style for promoting learner engagement in any study area.This finding can serve to remind practitioners that the 'speaker-centric' video style, featuring one or more human speakers as the main vehicle to deliver instructional content, should be used with discretion.

Table 1 .
Selected typologies of video styles.

Table 3 .
Summary of PCA results.

Table 4 .
Relationships between video styles and levels of engagement in arts and humanities MOOCs (n = 214).
Note: NS: not significant.

Table 5 .
Relationships between video styles and levels of engagement in computers, engineering, and mathematics MOOCs (n = 461).

Table 6 .
Relationships between video styles and levels of engagement in natural sciences MOOCs (n = 155).
Note: NS: not significant.

Table 7 .
Relationships between video styles and levels of engagement in social sciences MOOCs (n = 332).

Table 8 .
Relationships between video styles and levels of engagement in personal development MOOCs (n = 210).

Table 9 .
Summary of significant relationships between video styles and learner engagement across study areas.Wallis test results and significant Kruskal-Wallis test results with non-significant post-hoc multiple pairwise comparisons are not shown in the summary.

Table 10 .
Summary of hypotheses testing results.