Relationship between the Latent Profile of Online Socially Regulated Learning and Collaborative Learning Motivation

Abstract

Socially regulated learning (SoRL) is an important way to maintain the sustainable development of collaborative learning (CL). Usually, learners can achieve sustainable and high-quality SoRL with the intervention of teachers. To improve the appropriateness of the intervention, teachers need to clarify the profiles of SoRL to which learners belong, as well as the influence of collaborative learning motivation (CLM) and the relevant background variables. This study used three non-duplicate samples to provide evidence for the psychometric properties of the SoRL and CLM scales through item analysis, exploratory factor analysis (sample 1, n = 531), and confirmatory factor analysis (sample 2, n = 1278). The profiles of SoRL among university students were determined through latent profile analysis (sample 3, n = 909). This study identified three profiles of regulation (strong SoRL, progressive SoRL, and weak SoRL). The analysis of multivariate variance and multiple logistic regression methods further explored the differences in the dimensions of SoRL structures across different profiles and the extent to which CLM and background variables predicted profiles. The results showed that collaborative motivation (CM) and learning motivation (LM) were the predictors of learners’ transformation from a low regulation level to a medium regulation level. CM, LM, altruistic motivation, and major background were the predictors of learners’ transition from the medium regulation level to the high regulation level. Accordingly, teachers can provide learners with an appropriate external intervention to promote the improvement of SoRL. This study contributes to improving learners’ SoRL levels and promoting the sustainable development of education. In the future, the changing characteristics of learners’ SoRL profiles over time will be explored, and the application of learning process data will be strengthened.

1. Introduction

Collaborative learning (CL) has been regarded as an effective learning method all along, which plays a positive role in cultivating learners’ high-order thinking and promoting deep knowledge building [1]. However, CL faces many challenges at the practical level, such as uneven participation, lack of in-depth interaction, and so on [2,3]. These problems hinder the sustainable development of CL and make it difficult for learners to obtain ideal CL results [4]. In order to promote the sustainable and efficient progress of CL, learners need to regulate the cognitive, motivational, and emotional problems faced in the process of CL. In general, socially regulatory learning (SoRL) is a regulatory behavior that occurs at the group level [5]. SoRL is an advanced psychological function acquired by learners, and its acquisition requires appropriate intervention from teachers [6]. At present, the existing studies on SoRL have mainly focused on the revelation of process characteristics, but there is still a lack of empirical evidence on how to stimulate learners to produce high-quality SoRL [7].

Learners have differences in cognition, metacognition, motivation, emotion, and knowledge background. Therefore, their SoRL skill has diverse characteristics, and it is difficult for teachers to cope with the time and energy cost of implementing personalized intervention [8]. Therefore, this study focuses on how professional teachers obtain classified information on learners’ SoRL to provide more accurate personalized support for learners. Specifically, this study first determined the types of SoRL to which learners belong and then further analyzed the effects of collaborative learning motivation (CLM) and background variables on different profiles of SoRL. Finally, this study briefly analyzed how teachers provided corresponding intervention from the practical level.

SoRL is a type of learning in which learners understand group tasks, make group plans, choose and implement group task strategies, monitor group task progress, and regulate peer cognition, metacognition, motivation, emotion, and so on [9]. The understanding of SoRL focuses on several aspects. The first is the context information, which is the regulatory behavior between learners and peers in CL situations. The object of regulation extends from self to self and peers, and peers include both a certain peer at the individual level and several or all peers at the group level [10]. The second is the process attribute, and the occurrence of SoRL has obvious phased characteristics, including task understanding (TU), goal planning (GP), strategy application and regulation, and evaluation and reflection (ER) [11,12]. The third is to clarify the regulation content, including the cognition, metacognition, and motivational emotions of peers [10]. This is called the regulation of peer metacognition as shared metacognitive regulation. Only when motivation and emotion regulation are added to the regulatory content can regulatory learning be regarded as SoRL in a complete sense [13]. The timing of regulation usually occurs in two situations. One is in the process of maintaining normal CL, and the team members lead the collaborative process forward in a proactive way [14]. The other occurs in the process of CL when learners encounter difficulties and challenges, and it highlights the learners’ monitoring and controlling of the CL process [15].

SoRL can be analyzed from the perspective of process and competence view [16]. From a process perspective, SoRL is a type of learning in which learners use relevant learning strategies to advance the CL process toward deeper levels. From the perspective of competence view, researchers generally regard SoRL as learners’ competence to apply relevant regulatory strategies and determine the learners’ activation of SoRL by calculating the frequency of application of regulatory strategies [17]. Generally, the competence perspective can be seen as an analysis of learners’ offline SoRL level. However, only the quantitative statistics of these strategies cannot reveal the actual situation of SoRL. Socially shared metacognitive regulations (all-round-oriented and affirming regulation, social-oriented and elaborating regulation, and individual-oriented and passive regulation) were found through qualitative analysis of regulated learning [18]. At the same time, the relationship between each profile and learning performance was analyzed. Individual-oriented and passive regulators perform poorly in knowledge tests, all-round-oriented and affirming regulators perform moderately, and social-oriented and elaborating regulators perform well. Therefore, the qualitative analysis of SoRL can further illustrate the application effects of different regulatory strategies to promote the depth of CL. Table 1 provides an overview of the latent profiles related to SoRL in previous studies.

Learners assume two roles in CL. In general, learners are the winners of the benefits of CL; they are also the active managers of the CL who participate in the regulation and control of the CL process actively. However, it is not enough for learners to know about regulation strategies; they also need to be willing to activate and use them throughout the CL [19]. Therefore, CLM should be taken into consideration. Learning motivation (LM) is the source of motivation for learners to participate and persist in learning [20]. From the perspective of internal drive, LM is divided into cognitive drive, self-improvement drive, and accessory drive [21]. The cognitive drive is to meet the interest in learning and the need to master knowledge and skills, which belongs to internal motivation. Self-improvement of internal drive is to meet the needs of improving one’s own work competence, which belongs to external motivation. Accessory internal drive is the motivation to win the respect of teachers or peers and the sense of belonging to the group, so it also belongs to external motivation.

Table 1. Overview of latent profiles of SoRL.

Number	Profile	Specific Content	Ref.
1	Strong shared regulation	Learners get high scores on all shared regulatory variables.	[22]
	Progressive shared regulation	Learners get moderate scores on all shared regulatory variables.
	Weak shared regulation	Learners get lower scores on all shared regulatory variables.
2	Individual regulated learning	Individual learner regulates the learning activities of groups.	[23]
	Co-regulated learning	All learners regulate the learning activities of the group together.
3	Interrogative socially shared metacognitive regulation	The regulation is caused by the profound doubt of one learner, and the peer of this learner has a more comprehensive and universal response to it.	[24]
	Affirmative socially shared metacognitive regulation	After the learner states the fact, the peer confirms it.
	Interfering socially shared metacognitive regulation	Regulation is triggered by factual contributions and responded by limited responses that are meaningless.
	Progressive socially shared metacognitive regulation	Regulatory behavior is triggered by expressive statements, and learners also explain the regulatory behavior of peers.

Table 1. Overview of latent profiles of SoRL.

Number	Profile	Specific Content	Ref.
1	Strong shared regulation	Learners get high scores on all shared regulatory variables.	[22]
	Progressive shared regulation	Learners get moderate scores on all shared regulatory variables.
	Weak shared regulation	Learners get lower scores on all shared regulatory variables.
2	Individual regulated learning	Individual learner regulates the learning activities of groups.	[23]
	Co-regulated learning	All learners regulate the learning activities of the group together.
3	Interrogative socially shared metacognitive regulation	The regulation is caused by the profound doubt of one learner, and the peer of this learner has a more comprehensive and universal response to it.	[24]
	Affirmative socially shared metacognitive regulation	After the learner states the fact, the peer confirms it.
	Interfering socially shared metacognitive regulation	Regulation is triggered by factual contributions and responded by limited responses that are meaningless.
	Progressive socially shared metacognitive regulation	Regulatory behavior is triggered by expressive statements, and learners also explain the regulatory behavior of peers.

CLM is regarded as a compound form, including CM and LM [25]. The CM can be characterized by an accessory internal drive. LM can be characterized by the cognitive drive and self-improvement drive. During the interviews with learners participating in online CL, learners’ motivation for regulation also came from concerns about peer development. This kind of active assistance behavior had no utilitarian tendency, which showed a significant altruistic tendency [26]. Therefore, it is necessary to bring altruistic motivation (AM) into the framework of CLM. AM is a type of motivation for learners to help others without reward, which can influence learners’ participation behavior [27,28,29]. Based on the above analysis, AM, CM, and LM constitute a new category of CLM. Among them, CM is related to the accessory internal drive, and LM is related to the cognitive internal drive and self-improvement internal drive. AM belongs to the research category of social psychology, which is a type of shared motivation for learners to obtain internal satisfaction in the process of helping peers.

Learners’ sense of belonging to collaborative organizations helps them to better participate in CL activities. Learners’ need for a sense of belonging depends mostly on the social and emotional interaction between learners [30]. When social and emotional interaction is characterized by symmetrical participation and equal communication, learners can better participate in the management of the CL process and make better suggestions for the SoRL [31]. When the social and emotional interaction is out of balance, the learners feel the negative emotional interaction and participate in the learning process of social regulation in a passive or even rejected way [32]. There was a significant correlation between the shared metacognitive regulation model and LM. Among the three found shared metacognitive models, omni-directional and positive regulators, social-oriented and elaborating regulators have active LM, while individual-oriented and passive regulators have controlled LM [18].

The motivational and emotional barriers that hinder learners’ altruistic behavior should be removed to achieve effective cooperation [32]. Another study found that compared with other motivational factors, medical learners have the highest score of altruistic intention when providing a voluntary health care service [33]. However, the empirical evidence on SoRL and AM is still scarce. As an important part of CL, SoRL involves a wide range of cooperative and mutual assistance behaviors, which can reflect different degrees of prosocial characteristics. Therefore, learners’ participation in SoRL should be related to their own AM. In general, the research on motivation related to SoRL is limited, and most of the existing studies treat motivation as a dependent variable. At present, it is not clear to what extent the level of learners’ CLM is related to learners’ participation in SoRL.

There are different degrees of differences in learners’ age, gender, previous knowledge reserve, and online CL experience. Learners’ regulated learning gradually developed from internal self-regulated learning to external peer and SoRL with the increase in learners’ age [34]. SoRL is a process of monitoring and regulating CL tasks; the learner’s knowledge reserve also has an impact on learners’ participation in regulatory learning. However, there is a lack of research on SoRL with different background variables. It is necessary to carry out empirical research to find more sufficient empirical data.

Online context provides new time and space conditions for learners to carry out CL. In this context, there is a significant reduction in the amount of embodied information that supports interaction, which results in a variety of challenges that may emerge at any time [35]. Learners need to carry out SoRL activities at the group level to meet these challenges. At the same time, in order to maximize the effectiveness of regulation, teachers need to provide appropriate support according to the learners’ characteristics of SoRL. Latent profile analysis (LPA) is a person-centered approach that pays attention to the heterogeneity among individuals and provides support for the construction of a targeted intervention model. This study focuses on solving the following questions:

(1)

How do we develop and validate an online SoRL scale and a CLM scale?

(2)

What are the latent profile types of university students’ SoRL in an online collaborative context?

(3)

How do the CLM and individual background of university students predict their SoRL profiles?

2. Materials and Methods

2.1. Participants

In this study, a total of 3034 questionnaires were randomly sampled from a university in southwest China in the fall semester of 2022, and 2718 valid questionnaires were obtained with an effective rate of 89.58%. Among them, 531 questionnaires (sample 1) were used for exploratory factor analysis (EFA), and the sample of 1278 (sample 2) questionnaires was used for confirmatory factor analysis (CFA). Another 909 questionnaires (sample 3) were used to analyze the SoRL profile.

2.2. Instruments

The instruments consist of three parts.

(1)

Personal information. It was designed by the researchers, including gender, major, grade, and the number of semesters of CL.

(2)

Online SoRL scale. This study developed an online SoRL scale based on the existing SoRL scales and the grounded analysis of interviews with experts and learners with experience in online SoRL. The online SoRL scale includes 6 dimensions: TU, GP, process monitoring (PM), social help (SH), effort control (EC), and ER. Specifically, TU includes 3 sub-dimensions (understanding task content, understanding task goals, and understanding task strategies), GP includes 3 sub-dimensions (goal setting, time management, and role division), PM includes 4 sub-dimensions (monitoring TU, monitoring task progress, monitoring task quality and monitoring strategy effectiveness), SH includes 2 sub-dimensions (help-seeking strategy and help strategy), EC includes 2 sub-dimensions (self-will control and peer motivation), and ER includes 2 sub-dimensions (ER on collaboration process and ER on collaboration products).

(3)

This study designed measurement items based on the dimension connotations, including 5 items for TU (e.g., ‘I negotiate and understand the problems to be solved with my group members in online CL’), 4 items for GP (e.g., ‘I set phased goals with my group members and allocate time for each task in online CL’), 5 items for PM (e.g., ‘I ask questions to help my group members better understand the task requirements in online CL’), 3 items for social assistance (e.g., ‘I am willing to provide help to my group members in online CL’), 4 items for EC (e.g., ‘I actively participate in the group’s discussion activities in online CL’), and 5 items for ER (e.g., ‘I review the understanding of the task requirements with my group members, analyze the shortcomings and make improvement suggestions in online CL’). The Likert 5-point scale was used to score with 1 for “complete non-conformity” and 5 for “complete coincidence”. As the total score increased, the learners’ level of SoRL increased.

(4)

Online CLM scale. It was developed based on reference Pintrich [36] and Barnard et al. [37]. The online CLM scale includes 3 dimensions (CM, LM, AM). This study designed measurement items based on the dimension connotations, including 2 items for CM (e.g., ‘Achieving effective communication is my motivation for participating in online CL’), 3 items for LM (e.g., ‘Increasing my knowledge reserve is my motivation for participating in online CL’), and 2 items for AM (e.g., ‘In online CL, I am willing to interact with my group members’). The Likert 5-point scale was used to score with 1 for “complete non-conformity” and 5 for “complete coincidence”. The higher the total score was, the higher the learners’ level of CLM was.

2.3. Data Analysis Method

2.3.1. Project Analysis

The main purpose of project analysis is to examine the appropriateness of individual items on a scale. The project analysis included critical ratio method, item-total correlation analysis, and homogeneity test method, aiming to analyze the scale scientifically and objectively.

2.3.2. Factor Analysis

Factor analysis is a dimensionality reduction method that aggregates multiple variables into a few independent common factors, which usually include EFA and CFA. The purpose of EFA is to detect the structure of multivariate observation variables and reduce the dimension so that the variables with complex relationships can be condensed into several core factors [38]. EFA used principal component analysis for extraction and maximum variance rotation for rotation, selecting orthogonal rotation with Kaiser Standardization to explore the internal structure of the scale. During this process, it is necessary to judge whether the scale is suitable for EFA through sampling adequacy measures (Kaiser–Meyer–Olkin, KMO) and Bartlett value. The purpose of CFA is to test the consistency between the structural model constructed by the researchers and the actual data [39]. CFA is usually used to verify three kinds of model validity (construct validity, convergence validity, and discriminant validity).

2.3.3. LPA

The basic assumption of LPA is that the probability distribution of various responses of explicit variables can be explained by a small number of mutually exclusive potential variables, and the selection of explicit variables in each category has a specific tendency.

As a person-centered analysis method, LPA is based on probability or model technology, which is a variant of the traditional clustering analysis method. In the research of educational psychology, individual cognition and motivation are measured indirectly through sample explicit and measurable behavior performance. This method is better than traditional clustering analysis in detecting latent profiles [40].

The samples studied in LPA are composed of individuals from different potential categories, and the observation scores of individuals belonging to the same category on the same series of indicators are assumed to come from the same probability distribution. The latent profile model characterizes the distribution on a series of observed indicators and assumes that these indicators are normally distributed in each category. The latent profile model is used as a function of the probability of latent category K and the class-specific normal density.

The K-l and K profile models are analyzed and compared by The LPA method until the continuous model no longer shows better adaptability to the existing data [41,42]. This method uses many fitting indexes such as Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), Lo–Mendell–Rubin Likelihood Ratio Test (LMRT), and Information Entropy (Entropy) to evaluate the statistical model fitting. In general, if a model has a higher Entropy and a lower AIC and BIC, and the LMRT is significant, it means that the model has a higher degree of fitting.

2.4. Research Process

During the COVID-19 pandemic, most learners had gained online learning experience. In the post-pandemic period, online learning has become a normalized way of learning. Building on this foundation, teachers guide students to adopt online CL methods and engage in regular inquiry activities. In this study, a snowball sampling method was used to distribute questionnaires by contacting teachers who had set up online CL modules. This ensured that participants in the experiment all had experience in online CL to a certain extent.

The research process is described as follows. First, this study used the method of project analysis and factor analysis to develop the SoRL scale and the CLM scale. After that, the LPA method was used to classify the learners’ SoRL profiles, and the multiple logical regression method was used to establish the relationship between each subtype and the influencing factors to provide support for the progressive learners to a better type. The flow chart of the research process is shown in Figure 1. This study used Mplus 8.0, SPSS 25.0, and AMOS 24.0 to analyze the data.

3. Results

3.1. Project Analysis Results

This study first used the project analysis method to purify the data from the SoRL questionnaire. The results of the critical ratio test showed that the p-values of all items were less than 0.05, and the mean values were greater than three. Therefore, all items should be retained. The results of the correlation test showed that the Person’s correlation coefficients of all items were greater than 0.6, so no items needed to be deleted. The homogeneity test was conducted using Cronbach’s Alpha method; the Cronbach’s Alpha values of the scale as a whole and each dimension were above 0.9, and the corrected item-total correlation (CITC) coefficients were all greater than 0.6 (the standard should be greater than 0.4), which indicated that there was a good correlation between the analysis items.

This study used the same method to analyze the CLM questionnaire, and the results showed that all items of the CLM questionnaire could be retained, and there was a good correlation between the analysis items.

3.2. Factor Analysis Results

This study first conducted a factor analysis on the SoRL questionnaire. The EFA results showed that the KMO value was 0.966, which was greater than 0.6 and met the prerequisites of factor analysis. This indicated that the data could be used for factor analysis research. The data passed the Bartlett sphericity test (p < 0.05), which indicated that the research data were suitable for factor analysis. Through multiple analyses of the scale and the gradual deletion of items that did not conform to their respective factors, an initial measurement scale consisting of 23 items was formed. Among them, there are four items in TU, four items in GP, four items in PM, two items in SH, three items in EC, and six items in ER.

To test the validity of the above factor structure, this study conducted CFA on a scale that had previously undergone EFA. CFA required a different sample than EFA. Therefore, the sample for CFA was analyzed using the 1278 questionnaires. The software used was AMOS25.0, and model fit was analyzed using the absolute fit index, incremental fit index, and parsimony fit index. The initial analysis results were CMIN/DF = 5.407, RMSEA = 0.096, RMR = 0.011, CFI = 0.953, and TLI = 0.943. For indicators that did not meet the fitting conditions, the Bollen–Stine Bootstrap method was used for correction. The modified model fit indices were CMIN/DF = 3.09, RMSEA = 0.07, RMR = 0.011, CFI = 0.98, and TLI = 0.968. Through the above analysis, this study ultimately formed a SoRL scale with a certain scientific and reasonable nature. The Cronbach’s α of the scale was 0.965 (sample 2, SPSS statistics). The online SoRL questionnaire is shown in Table S1.

Subsequently, this study used the same method to conduct factor analysis on the CLM scale. The results showed that the KMO value was greater than 0.6, and the data passed the Bartlett sphericity test (p < 0.05), which indicated that the data were suitable for factor analysis. The EFA revealed that the factor analysis results were generally consistent with the research concept, covering three dimensions of the theoretical concept. However, the fitting results of the scale in CFA were not ideal (CMIN/DF = 7.271; RMSEA = 0.083; RMR = 0.006; CFI = 0.992; TLI = 0.984). The factor loading of the third question in the CLM subscale was relatively low. After examining the modification indices and item descriptions of the subscale, the item was deleted, and the model was re-run. The fitting of the model was improved to some extent (CMIN/DF = 3.12; RMSEA = 0.071; RMR = 0.006; CFI = 0.994; TLI = 0.988). The final CLM scale consisted of three subscales: CM included three items, and LM and AM each contained two items. The Cronbach’s α of the scale was 0.983 (sample 2, SPSS statistics). The collaborative LM questionnaire is shown in Table S2.

3.3. Statistical Results of the SoRL Variables

Table 2. The means, standard deviations, and correlation coefficients of the variables of SoRL and CLM.

Item	M	SD	TU	GP	PM	SH	EC	ER	CM	LM	AM
TU	4.263	0.834	1
GP	4.253	0.830	0.927 **	1
PM	4.212	0.826	0.860 **	0.906 **	1
SH	4.160	0.884	0.760 **	0.815 **	0.838 **	1
EC	4.252	0.790	0.851 **	0.883 **	0.901 **	0.835 **	1
ER	4.246	0.793	0.854 **	0.882 **	0.894 **	0.817 **	0.923 **	1
CM	4.259	0.789	0.834 **	0.855 **	0.859 **	0.807 **	0.883 **	0.922 **	1
LM	4.273	0.792	0.823 **	0.840 **	0.843 **	0.792 **	0.880 **	0.928 **	0.930 **	1
AM	4.274	0.792	0.828 **	0.845 **	0.847 **	0.804 **	0.885 **	0.911 **	0.924 **	0.918 **	1

Note: ** p < 0.01.

presents the means, standard deviations, and Pearson correlation coefficients for the SoRL variables. In terms of means, TU, goal setting, and EC received higher scores, while PM, SH, and ER received lower scores. The distribution of standard deviations was similar, with a larger range of values for SH. The SoRL strategies showed a strong correlation with their CLM (r > 0.7). This correlation indicated an internal association between SoRL and CLM.

3.4. Latent Profile of SoRL among University Students

To differentiate the different types of SoRL levels among university students, a latent profile model was established using six dimensions (TU, GP, PM, SH, EC, ER) as indicators. Starting with a baseline model with a category number of 1, an LPA was conducted on SoRL to fit the model.

This study used fitting indices and standards to select the best profile model. The data were sequentially input into codes with different numbers of categories for analysis. The fitting indices of potential profile models for categories 1–5 are shown in

Table 3. LPA of university students’ SoRL.

Clusters	BIC	AIC	ABIC	Entropy	LMR (p)	BLRT (p)	Group Size
1 profile	14,055.824	13,998.076	14,017.714	——	——	——	909
2 profiles	10,173.527	10,082.093	10,113.186	0.982	0.0183	0.0171	454/455
3 profiles	7105.187	6980.066	7022.615	0.99	0.0484	0.0462	324/147/438
4 profiles	5008.100	4849.292	4903.296	0.993	0.2205	0.2154	17/312/142/438
5 profiles	4328.880	4136.387	4201.846	0.998	0.7560	0.7534	17/127/295/442/28

. Overall, from the perspective of model fitting, the values of AIC, BIC, and Akaike’s Bayesian Information Criterion (ABIC) decreased as the number of categories increased. From the perspective of model classification accuracy, the entropy values for categories four and five were the highest. In terms of model fit significance, the BLRT values for categories two and three were significant, but the AIC and BIC values for category three were smaller. The LMRT showed that the three-profile model was more suitable for the data than the two-profile model, and the three-profile model had lower AIC and BIC values and higher entropy values. In addition, the sample size of the groups and the theoretical interpretability were examined, and the result of the three profiles was optimal.

The proportion of the three profiles of SoRL in the total and the application of each regulatory focus in the three profiles are shown in Figure 2. The first regulation type scored in a middle state across all dimensions, with 324 learners belonging to this category (accounting for 35.64%), which was defined as the group of progressive SoRL. Learners in the second regulation type scored low across all dimensions, with 147 learners belonging to this category (accounting for 16.17%), which was defined as the group of weak SoRL. Learners in the third regulation type scored high across all dimensions, with 438 learners belonging to this category (accounting for 48.18%), which was defined as the group of strong SoRL.

In the study of SoRL, researchers generally characterized regulatory strategies as the regulation focus. The status of each regulation focus in the three profiles is shown in

Table 4. Specific conditions of regulation focus in three profiles.

Regulation Focus	Progressive SoRL (n = 324)	Weak SoRL (n = 147)	Strong SoRL (n = 438)
	M (SD)	M (SD)	M (SD)
TU	3.97 ± 0.23	2.81 ± 0.59	4.97 ± 0.11
PS	3.96 ± 0.29	2.88 ± 0.69	4.93 ± 0.23
MS	3.91 ± 0.39	2.98 ± 0.69	4.85 ± 0.40
SH	3.86 ± 0.53	3.02 ± 0.78	4.76 ± 0.59
EC	3.97 ± 0.35	3.10 ± 0.79	4.85 ± 0.36
ER	3.93 ± 0.38	3.10 ± 0.74	4.86 ± 0.34

. Further analysis revealed significant differences between the SoRL profiles. Upon further analysis of the scores for specific regulation strategies within each profile, learners in the progressive and strong profiles scored higher on TU and GP, while learners in the weak SoRL profile scored lower in these aspects. This suggested that the level of TU and GP had a significant impact on the quality of SoRL.

3.5. Predicting Individual Data for SoRL

Question 3 was about to what extent CLM and learners’ background characteristics predicted the profile of SoRL they belong to. CM, LM, and AM were used as predictors of CLM. Gender, major background, grade, and the number of online CL semesters were used as background predictors. On this basis, multiple regression analysis was conducted.

This study used the progressive SoRL profile as a reference profile and compared the weak SoRL profile with the progressive SoRL profile (

Table 5. Multivariate logistic analysis of the factors influencing the transformation from weak SoRL to progressive SoRL.

Weak SoRL Profile vs. Progressive SoRL Profile (95% CI for OR)
Item	Parameter	B (SE)	z	Wald χ2	OR	Lower	Upper
Background characteristics	Gender	−0.138 (0.261)	−0.531	0.282	0.871	0.522	1.452
	Major	0.086 (0.251)	0.341	0.116	1.089	0.666	1.782
	Grade	0.065 (0.106)	0.612	0.374	1.067	0.866	1.314
	CL experience	−0.060 (0.150)	−0.397	0.157	0.942	0.702	1.264
CLM	CM	−1.288 (0.432) **	−2.984	8.903	0.276	0.118	0.643
	LM	−1.349 (0.411) **	−3.283	10.779	0.259	0.116	0.581
	AM	−0.359 (0.390)	−0.920	0.846	0.699	0.325	1.5

Note: ** p < 0.01.

). Both CM and LM showed significant negative effects. Specifically, as indicated by the odds ratios (OR), the likelihood of learners transitioning from the weak SoRL to the progressive SoRL level increased by 0.276 and 0.259 times with each unit increase in CM and LM, respectively. Background variables and AM did not show any significant effects on the profile outcomes.

This study also used the progressive SoRL profile as a reference profile and compared the strong SoRL with the progressive SoRL (

Table 6. Multivariate logistic analysis of the factors influencing the transformation from progressive SoRL to strong SoRL.

Strong SoRL Profile vs. Progressive SoRL Profile (95% CI for OR)
Item	Parameter	B (SE)	z	Wald χ2	OR	Lower	Upper
Background characteristics	Gender	−0.151 (0.252)	−0.600	0.360	0.860	0.524	1.409
	Major	0.628 (0.253) **	2.480	6.150	1.874	1.141	3.078
	Grade	−0.089 (0.107)	−0.832	0.693	0.915	0.743	1.128
	CL experience	0.033 (0.137)	0.242	0.058	1.034	0.79	1.353
CLM	CM	1.752 (0.394) **	4.451	19.808	5.767	2.666	12.475
	LM	1.023 (0.370) **	2.762	7.628	2.781	1.346	5.747
	AM	1.762 (0.369) **	4.777	22.823	5.827	2.827	12.008

Note: ** p < 0.01.

). The discipline background in the background variables showed a significant positive effect. The CLM, including CM, LM, and AM, showed a significant positive effect. Specifically, the increase in discipline background from the middle regulatory level group to the high regulatory level group was 1.874 times after each unit increase. The possibility of the learners’ CM, LM, and AM changing from the progressive SoRL group to the strong SoRL group increased by 5.767 times, 2.781 times, and 5.827 times with each increase in one unit. Gender, grade, and the number of semesters participating in CL did not show a significant effect on the profile results.

4. Discussion

The goal of this study was to develop an online SoRL and CLM scale and explore the latent profiles of university students’ SoRL in an online collaborative context and its relationship with CLM and background variables. This study further discussed the limitations of the study and the direction of future research, which provided relevant strategic guidance on teachers’ intervention at the practical level.

4.1. Analysis of SoRL and CLM Scale

This study set out to develop a self-report measure for assessing SoRL and CLM in online collaborative contexts. A six-factor structure of SoRL and a three-factor structure of CLM in an online context were developed and tested. The results of EFA supported our theoretical conceptualization, which included (1) TU, (2) GP, (3) PM, (4) EC, (5) social assistance, and (6) ER. CLM included (1) CM, (2) LM, and (3) AM. Using a non-repeated sample, CFA was conducted to further confirm the factor structures of the two scales and provide evidence of their reliability and validity.

The factors found in the EFA of SoRL and CLM scale were validated in the CFA. In addition, the newly constructed factor structure in SoRL expands the connotation of existing scales. For example, the TU factor was not covered in the SoRL scale developed by Su et al. [43], but it played an important role in the early stages of CL activities. This was because learners only generated a series of CL behaviors later if they initially reached a consensus on the collaborative task. At the same time, in the co-regulation scale developed by Olakanmi [44], only monitoring elements were included, and ER elements were not involved. Normally, monitoring was a regulatory behavior after comparing with the established goals, which usually occurred in the process of CL. However, if monitoring behavior was carried out at the end of CL activities, then the management of CL activities tended to be finalistic, so it was more appropriate to use ER currently.

This study added the element of AM to the previous research on CLM. Through EFA and CFA, AM constituted an element of CLM. The reason for this was that both collaborative and LM were seen as types of motivation that stimulate learners’ own participation intentions, while AM was a type of motivation that stimulates peers’ participation intentions. CL was a complex interaction between learners. When learners were concerned about their peers, they better participated in the management of the CL process. Therefore, AM was an indispensable part of the framework of CLM.

4.2. Analysis of SoRL Profiles

The results of LPA indicated that university students’ SoRL was classified into three profiles (strong SoRL, progressive SoRL, weak SoRL) in the online CL context. Learners’ social and emotional interactions during CL also exhibited clustering characteristics, which were divided into three types (negative, neutral, and diverse) [45]. Learners’ self-regulated learning in blended learning environments was classified into three profiles (high, low, and moderate) Vanslambrouck et al. [46]. The profiles identified in this study reflected the frequency of learners’ use of SoRL strategies. In terms of the proportion of individuals in each of the three profiles (strong, progressive, weak), the number of individuals in the weak SoRL profile was the smallest, which indicated that university students had a certain level of SoRL competence. However, since the data used for profile analysis came from learners’ subjective reports in an offline state, it was possible that the reported level of SoRL was higher than the actual level.

There were differences in the application frequency of regulation focus in different profiles of SoRL. Learners with progressive and strong regulation level profiles tended to use the regulation focus, such as TU, goal setting, and PM. The weak SoRL learners had the lowest frequency of application in TU and goal setting. This showed that TU and goal setting were important factors that determine the learning quality of SoRL. Some researchers have found a positive correlation between goal setting and academic achievement for learners [47,48]. Goal setting was critical to achieving good collaboration performance [49]. Learners had different cognitive, metacognitive, motivational, and emotional preferences. Individuals needed to regulate their own goal direction to make the cooperative goals of group members as consistent as possible for further forming group cognition for task processing when learners changed from an individual state to group collaboration [50]. In addition, PM was the regulatory focus at the strong SoRL and an important factor that determined the quality of CL [51].

4.3. Predictive Role of CLM and Background Variables (RQ2)

Question 3 was to explore the influence of learners’ background variables and CLM on the latent profile of SoRL. There had been a rich accumulation of related research on LM and SoRL, and a universal consensus had been reached on the relationship between them. In general, LM had a positive impact on the improvement of academic performance. In terms of the relationship between LM and SoRL, the increase in learners’ motivation regulation can improve the academic performance of people with low self-regulated learning competence [52]. However, the study found that there was no significant correlation between learners’ motivation level and learners’ participation in shared metacognitive regulation [32,53].

This study focused on the relationship between the SoRL profile and CLM. CM and LM influenced learners’ transformation from weak SoRL to progressive SoRL, and the influence of CM was strong. This result further revealed the relationship between shared metacognitive regulation profiles and learners’ LM. There was a close relationship between learners’ shared metacognitive regulation profiles and LM. Comprehensive and affirmative regulators and social-oriented collaborative learners were both affected by autonomous LM, and individual and passive learners were mainly affected by controlled motivation [18]. The reason for this difference may be that the enthusiasm of learners to participate in the management of the CL process was affected by peer relationships [54]. When learners gain the respect of teachers in the CL process and form a better collaborative relationship with their peers, they may invest more time and energy to participate in the management of CL.

CM, LM, and AM influenced learners’ transition from progressive SoRL to strong SoRL, and AM had the greatest influence. This showed that when learners’ SoRL levels shifted from medium to strong, learners needed to strengthen their investment in AM because of higher levels of CM and LM.

The advanced psychological function of learners first appeared at the social level and then appeared at the psychological level [55]. Hence, learners should not only be concerned about their own benefits in the process of CL but also actively participate in the interaction with their peers and constantly promote the transfer of regulatory competence from the social level to the self level. In addition, the learner’s major background only affected the change from progressive SoRL to strong SoRL. Gender, grade, and the number of semesters participating in CL had no effect on the profile results.

4.4. Limitations and Future Research

Although this study provided insights into the SoRL profile and the mechanism of influencing factors, there were still several limitations that need to be addressed. First, this study used a cross-sectional method to eliminate the characteristics of SoRL and CLM changing over time. In fact, learners’ participation in SoRL and CLM were dynamic variables in the process of online CL. Therefore, it was necessary to explore the changing characteristics of learners’ SoRL profiles over time. Secondly, the data analyzed in this study were mainly in the form of a questionnaire, which was subjective and different from the SoRL characteristics shown by learners in the process of collaborative task processing. Future research should strengthen the application of learning process data. Thirdly, the SoRL questionnaire adopted in this study mainly focused on cognitive and metacognitive regulation strategies, but this did not include motivation and emotion regulation. The complex relationship between learners’ motivation and motivation regulation strategies needs to be further clarified. Incorporating additional controls for variables that influence the relationship between CLM, background variables, and SoRL profiles will be considered. Therefore, future studies will delve into these areas.

4.5. Practical Inspiration

The results of this study can provide support for improving learners’ SoRL levels. Learners’ SoRL presents different profile features. Teachers should adopt different strategies to intervene by analyzing the current profile to which learners belong when providing intervention support.

When promoting the transformation of learners’ SoRL from weak SoRL to progressive SoRL, teachers should focus on strengthening learners’ CM and increasing learners’ sense of belonging to the team. For example, teachers should establish a collaborative atmosphere of respect, equality, and tolerance, encourage learners to speak freely in the collaborative process, and provide sufficient learning resources for the collaborative process to facilitate learners’ discussion of learning resources [56]. In addition, teachers should guide collaborative groups to establish a promotive goal structure. In the early stages of CL, teachers should guide learners to construct a promoting relationship between individual learning goals and group learning goals, avoiding students’ excessive focus on individual learning goals while neglecting group learning goals. Therefore, in calculating academic performance, teachers should cover both individual and group performance and appropriately increase the proportion of group performance in the total score.

When promoting the transformation of learners’ SoRL from progressive SoRL to strong SoRL, teachers should focus on strengthening AM. Altruism involves learners’ concern and support for their peers, and it is the type of motivation that benefits the well-being of peers from the perspective of learners. Therefore, teachers encourage learners to strengthen their guidance and care for peers by adding rewards. Teachers can confirm the specific reward recipients in two ways: first, group members’ recommendations; second, in-depth interviews with collaborative groups to identify learners who have contributed to guiding and caring for their peers in the collaborative group. On this basis, teachers can give relevant rewards to those who have made outstanding contributions, trying to increase the proportion of spiritual rewards as much as possible. In addition, teachers should optimize the weight of evaluation indicators for course performance, pay attention to learners’ management of CL processes, and increase the proportion of students’ care and guidance for their peers. In terms of obtaining evaluation data, teachers can obtain detailed information about peer evaluations through surveys or interviews.

5. Conclusions

SoRL was the key to achieving high-quality CL. This study used the person-centered approach to prove the differences in SoRL level of university students in the context of online collaboration, and the conclusions include the following three aspects.

(1)

The online SoRL scale includes six dimensions: TU, GP, PM, SH, EC, and ER. The online CLM scale includes three dimensions: CM, LM, and AM.

(2)

There are three profiles of SoRL (strong SoRL, progressive SoRL, and weak SoRL). In terms of the proportion of personnel, the weak SoRL accounted for the least, which indicated that the current SoRL level of university students is mainly in the upper-middle level.

(3)

CM and LM were predictors of learners’ transformation from weak SoRL to progressive SoRL. CM, LM, AM, and major backgrounds were the predictors of learners’ transition from progressive SoRL to strong SoRL. The results were helpful to have a further understanding of the predictive role of CLM and background variables.

(4)

In order to promote the transformation of learners from weak SoRL to strong SoRL, teachers should provide appropriate external interventions such as creating a collaborative atmosphere, establishing a promotive goal structure, encouraging learners to strengthen their guidance and care for peers, and increasing the proportion of evaluation of caring and guiding peers in the overall evaluation index.