The SRL, initially proposed by Zimmerman (1986) as metacognitively, motivationally, and behaviorally active participation by learners in their own learning process, has since been defined as “the ways that learners systematically activate and sustain their cognitions, motivations, behaviors, and affects, toward the attainment of their goals” (Schunk and Greene, 2018, p. 1). This multidimensional and goal-oriented concept emphasizes learners’ active participation in regulating their own learning and the systematic and dynamic interaction of cognitive, metacognitive, motivational, affective, and environmental factors in this process (Teng and Zhang, 2021).

The process-oriented and goal-directed nature of L2 writing, supported by cognitive writing models (Flower and Hayes, 1981), highlights the relevance of the SRL perspective to the writing process. This approach recognizes that self-regulated writing processes involve the integration of cognitive, metacognitive, affective, and environmental factors (Panadero, 2017). Large-scale questionnaire studies have shown the value of SRL strategy use in relation to L2 writing performance (Teng and Zhang, 2016; Csizér and Tankó, 2017). Such studies have revealed the individual differences that correlate with SRL strategy use, such as personality (Jackson and Park, 2020), self-efficacy (Teng, 2021), motivational belief (Bai and Wang, 2021), and mindset (Bai and Wang, 2020). SRL strategy intervention studies have further revealed the effectiveness of explicit cognitive and metacognitive strategy use on L2 writing achievements (Hu and Gao, 2017; Chamot and Harris, 2019; Teng and Zhang, 2020). The longitudinal development of self-regulated writing strategies was also examined in mixed-methods research by Sasaki et al. (2018). They investigated the interaction of self-regulated writing strategies (e.g., global planning, local planning, and L1–L2 translation) with cognitive, affective, and environmental factors in L2 writing. Nonetheless, learners’ SRL strategy use in finely focused areas of the L2 writing process remains under-researched (Andrade and Evans, 2013; Xiao and Yang, 2019).

Based on a paradigm shift from “what writers write” to “how writers write,” process-oriented writing (Zamel, 1976) views writing as a cyclical process based on rounds of revisions driven by feedback from peers or teachers, which aims to improve text quality (Kroll, 2001). During revision in response to feedback, learners actively participate in decision-making processes by employing several strategies and meta-strategies for text improvement. For example, the study of Yu and Lee (2016) reveals that Chinese EFL learners have employed five major strategies while engaging in peer feedback activities (using L1, employing L2 writing criteria, adopting rules of group activity, seeking help from teachers, and playing different roles) to facilitate their peer interaction and successful revision learning. Writers’ active participation and dynamic use of strategies associated with feedback and revision to regulate their own writing are compatible with the concept of SRL, which emphasizes how learners effectively control their own learning process: “To develop SRL, learners must be actively engaged in the decision-making processes related to revision and editing” (Andrade and Evans, 2013, p. 50). Hattie and Timperley (2007, p. 94), in their model using feedback to enhance learning, proposed that self-regulation “can lead to seeking, accepting, and accommodating feedback information.” Studies have shown that self-regulated learners can interpret and evaluate feedback from different sources to facilitate their own learning goals (Lam, 2015).

The SRL strategies learners deploying to handle feedback from multiple sources are essential for developing self-regulated writers. Studies have revealed a significant correlation between feedback handling and revising strategies during the various stages of feedback revision and writing competence (Bai, 2015; Teng and Zhang, 2016; Bai and Wang, 2020). For example, Teng and Zhang (2016, p. 681) developed and validated the four categories of writing strategies for an SRL questionnaire in the Chinese EFL writing context, namely, cognitive strategies (e.g., text processing and course memory strategies) refer to skills students use to process the information or knowledge in completing a task; metacognitive strategies (e.g., idea planning and goal-oriented monitoring and evaluation) refer to the skills used to control and regulate learners’ own cognition and the cognitive resources they can apply to meet the demands of particular tasks; social behavior strategies (e.g., feedback handling and peer learning) involve individuals’ attempts to control their learning behavior under the influence of contextual and environmental aspects; motivational strategies (e.g., motivational self-talk, interest enhancement, and emotional control) refer to the procedure or thoughts that students apply purposefully to sustain or increase their willingness to engage in a task. Feedback handling, open to peer or teacher feedback and improving students’ own writings accordingly, is one of the six self-regulated writing strategies that significantly predicts EFL writing proficiency. Bai’s (2015) study on writing strategy intervention used an experimental research design and found that three strategies, i.e., text-generating, feedback handling, and revising, significantly contribute to the improvement of writing competence after intervention. Feedback handling and revising strategies were embedded in the whole writing process of the writing strategy-based instruction lessons, from setting writing goals to revising the essay and getting and responding to feedback. Subsequently, Bai (2018) examined self-regulated writing strategies among 32 Singapore primary pupils based on the think-aloud protocol and found that revising strategies were frequently used, such as revising mechanics, revising ideas at clause/sentence level, revising during drafting, revising after drafting, and revising by deletion and substitution.

However, most of these studies have examined self-regulated writing strategies throughout the writing process (Flower and Hayes, 1981), incorporating planning, organizing, and monitoring in addition to revising (De Silva and Graham, 2015; Hwang and Lee, 2017; Sun and Wang, 2020). Revising strategies, in these studies, are categorized merely as cognitive strategies about how to revise (e.g., revising mechanics; making lexical, morphological, and syntactic changes; revising ideas at the clause/sentence level) rather than strategies and meta-strategies for making revising decisions based on feedback from external sources. Feedback handling strategies are simple statements aimed at feedback and revision activities in general (e.g., I am open to teacher feedback on my writing; I read peers’ feedback) rather than contextualized and fine-grained strategies used in response to feedback from different sources.

The notion of learner engagement with feedback considers both feedback characteristics and feedback uptake and facilitates our comprehension of feedback revision processes (Han, 2017; Suzuki et al., 2019). Consensus has been reached regarding learners’ multidimensional engagement with multiple feedback sources from cognitive, affective, and behavioral perspectives (Zhang and Hyland, 2018). Case studies that closely examined learner engagement with feedback have reported the use of cognitive and metacognitive strategies (Han, 2017; Tian and Zhou, 2020). When revising based on feedback, learners’ cognitive strategies utilize, e.g., linguistic, rhetorical, and discourse knowledge to revise a text with reference to writing lectures attended. Metacognitive strategies refer to specific idea-generating behaviors before revising, which aim at directing and monitoring revision activities. When engaging in feedback revision activities, learners might also need to ask for help from others and constantly regulate their motivation and maintain a positive frame of mind to complete the feedback revision tasks. Considering the multidimensional nature of learner engagement with feedback and revision (Zhang and Hyland, 2018), revising based on feedback likely entails multidimensional SRL strategy use (Teng and Zhang, 2021).

Furthermore, evidence from distance language learning supports the role of autonomy (Murphy, 2005). This implies that learners in online writing contexts where multiple forms of feedback are conveyed to learners through online platforms might need to devote more effort to planning, reflecting, monitoring, and evaluating during feedback revision processes. Previous studies in online contexts have examined either student and teacher perceptions of online writing courses having adopted electronic feedback (Tai et al., 2015; Litterio, 2018), or learner engagement with different feedback sources (Tian and Zhou, 2020). Those studies (e.g., Jansen et al., 2017) examining SRL in online contexts have mainly advocated the importance of SRL in L2 online learning, consisting of metacognitive skills, environmental structuring, help-seeking, time management, and persistence. Few studies have examined SRL strategy use in the feedback revision stage of the L2 writing process in an online writing context.

This study aims to fill this gap by examining SRL strategy use when revising based on automated, peer, and teacher feedback in an online EFL writing context through a mixed-methods approach. It intends to understand fine-grained strategy use and the driving force behind the feedback uptake decision-making process. The following research questions were examined.

College English Writing II is an online English writing course at one of the top five universities in Beijing, China. The course is one of the three online writing courses in the target university and is an optional course for undergraduates. This course, run every spring term, aims to foster learners’ abilities in expository and argumentative writing. Learners are supposed to watch 16 online video lectures (40 min each), attend three face-to-face lectures, write three essay assignments, and complete three feedback revision tasks for each essay assignment. Learners who successfully complete all tasks gain course credits.

The three-stage feedback revision framework was specifically designed for this online English writing course to help learners improve their writing through multiple revisions. “Three-stage” refers to three different feedback sources, namely, an automated writing evaluation (AWE) program, peers, and the teacher. Each assignment topic undergoes the three-stage feedback revision procedure before a final essay is submitted. Learners can receive multiple feedback successively from the three sources and revise their essay drafts accordingly. AWE feedback is given by Pigai¹, a widely used online AWE program in China (Bai and Wang, 2018; Tian and Zhou, 2020). Peer feedback pairs are randomly allocated by the online writing program, and teacher feedback is also conveyed online.

For this study, the second of the three essay assignment topics, an argumentative essay topic, was chosen as the target topic. By that time, learners were familiar with the feedback revision procedure and could devote themselves to the task. The assignment was as follows:

A convenience sample of 52 first-year students enrolled in the online English writing course was invited to participate in the study. All participants were non-English majors from a wide range of subjects, such as philosophy, international relations, journalism, finance, business, and French. However, only 36 of them voluntarily completed all the three questionnaires at three different feedback-revision stages.

In addition to complete the questionnaires, four of the 36 participants (Table 1) voluntarily agreed upon invitation to follow think-aloud protocols and respond to stimulated recall interviews. Three of these four students were women, and they came from various subject backgrounds. An English Placement Test, administered to all first-year students upon entering the university, was available for use as a reference for their general proficiency level.

We adopted a mixed-methods approach to investigate which self-regulated writing strategies are used and how they are used when learners revise in response to multiple feedback sources from AWE, peers, and the teacher over a month. Both retrospective and introspective data were collected (Figure 1) to capture learners’ spontaneous and reported writing strategy use during the three-stage feedback revision process. In each feedback revision cycle, the four selected participants followed the think-aloud protocol. At the end of each feedback revision cycle, all 36 learners completed the questionnaire. After all feedback revision stages were completed, the four selected learners responded to a stimulated recall interview. Therefore, in total, the data comprised 108 questionnaires (36 × 3), 4 stimulated recall interview transcripts, 12 sets of think-aloud data (4 × 3), 12 rounds of feedback from the 3 feedback sources (4 × 3), and 24 drafts and submitted essays from the 4 participants during the 3 feedback revision stages (4 × 3 × 2).

Data analysis involved quantitative analysis of questionnaires, content analysis of think-aloud and stimulated recall interview data, and textual analysis of 12 rounds of feedback and 24 essays. Questionnaire data were statistically analyzed using SPSS. The mean frequency of each of the four strategy categories was calculated for each case. Based on Kolmogorov–Smirnov (K–S) tests of normality (Table 3), parametric tests, such as repeated measures ANOVA, were conducted to examine strategy use differences across feedback sources.

After being transcribed verbatim by the third author, the think-aloud data were analyzed qualitatively with reference to self-regulated writing strategies. First, all authors conducted open coding to identify strategy categories and then matched them with the strategy category list in the questionnaire. A combination of bottom-up and top–down data mining methods was used to precisely identify feedback revision strategies in all stages, together with iterative identification. All disagreements were resolved among the authors based on the repeated reading of transcripts and rounds of discussions. Stimulated recall interview data were transcribed verbatim before being analyzed qualitatively to triangulate with the self-regulated writing strategies reported in the questionnaire and think-aloud data.

Textual analysis of feedback and revisions was based on a coding scheme (Table 4) adapted from Faigley and Witt (1981) and Allen and Mills (2016); the scheme was also adopted by Tian and Zhou (2020), who used it to calculate feedback and feedback uptake rates. In this approach, feedback that did and did not suggest changes in meaning was classified as meaning- and surface-level feedback, respectively. Feedback from AWE, peers, and teachers was coded accordingly. Revisions were first conducted by comparing draft and submitted essays at each of the three feedback revision cycles and stages. Subsequently, revisions were compared with feedback at each stage to differentiate feedback according to whether it resulted in revisions. Incorporated feedback was counted as feedback uptake; uptake rate was then calculated by dividing total feedback by feedback uptake.

Thirty-six learners completed three questionnaires pertinent to the three feedback sources, retrospectively reporting their strategy use when revising. Table 5 shows that all learners employed all four strategies across the three feedback revision stages to a large extent. Frequencies ranged from 3.51 (social strategy based on peer feedback) to 4.44 (motivational strategy based on teacher feedback) along a 5-point Likert-type scale ranging from 1 (not at all true of me) to 5 (very true of me).

Qualitative analysis of both think-aloud and stimulated recall interview data showed that the four analyzed cases, regardless of their proficiency levels, employed all strategies to regulate their feedback revision activity. This is in accordance with the quantitative findings on strategy use and actual revision practice.

Table 5 shows that cognitive and motivational strategies were used more frequently than metacognitive and social strategies when revising, regardless of the feedback source. The ANOVA confirms significant strategy-use differences according to feedback sources, AWE: F(3,105) = 9.331, p = 0.000, η² = 0.210; peer: F(3,105) = 4.645, p < 0.01, η² = 0.117; teacher: F(3,105) = 15.354, p = 0.000, η² = 0.305.

Table 5 shows that more cognitive strategies were used when revising based on AWE feedback than the other feedback sources, while more motivational strategies were used when revising based on teacher feedback. A repeated measures ANOVA showed that cognitive strategy use was not significantly affected by the three feedback sources, F(2,70) = 2.133, p > 0.05. However, post hoc tests showed a significant difference in cognitive strategy use between AWE and peer feedback sources, F(1,35) = 5.096, p = 0.03, with a medium effect size (η² = 0.13). Compared with peer feedback (mean = 3.97), learners used significantly more cognitive strategies with AWE feedback (mean = 4.16). Metacognitive strategy use was the highest for teacher feedback (mean = 3.91). However, a repeated measures ANOVA showed that metacognitive strategy use was not significantly affected by the three sources, i.e., F(2,70) = 0.475, p > 0.05, nor was social strategy use, F(2,70) = 0.456, p > 0.05. Nevertheless, motivational strategy use was found to be significantly affected by the three sources, F(2,70) = 12.178, p = 0.000, with a large effect size (η² = 0.26). The post hoc tests also showed significant differences in motivational strategy use between AWE and teacher feedback (p = 0.003) and between peer and teacher feedback (p = 0.000).

Textual analysis of feedback and feedback uptake revealed three feedback quantities for the four participant cases and their feedback uptake rates (Table 6). Notably, AWE feedback is the most frequent overall (72.3–88.5%), although this is mostly at the surface level (over 92.7%), and produces a low feedback uptake rate (11.1–33.3%). Conversely, teacher feedback was less frequent (7–23.1%) but was all aimed at the meaning level (100%) and had the highest feedback uptake rate (100%). Peer feedback was the least frequent (0–16.3%), was mostly at the meaning level, and had the greatest variation in uptake rate (0–55.6%).

The feedback quantity and uptake of the four cases (Table 6), together with the strategy use of the 36 participants when revising based on three feedback sources (Table 5), revealed an overall tendency of feedback features and the corresponding strategy use for different feedback sources. The high quantity of surface-level AWE feedback seems to demand more cognitive strategies during subsequent revisions, and the high teacher feedback uptake rate at the meaning level appears to elicit more motivational strategy use during revision. Hence, feedback type, quantity, and uptake seem to be associated with self-regulated writing strategy use.

Think-aloud data and interviews indicated high cognitive strategy use in the four cases when revising upon AWE feedback, where they carefully checked each piece of feedback in relation to their original essays and revised accordingly. The high amount of AWE feedback might have led to excessive text processing, which referred to what learners had acquired in both online and offline writing lectures. For example, Li, a low-proficiency student, reported on her decision-making and revision processes in her think-aloud data: “It [Pigai] said ‘basic’ is an adjective but was wrongly used by me as a noun. Yes, it is right. Here, it should be ‘basis’.” Stimulated recall interviews corresponded to the think-aloud data. For example, for the above AWE feedback, Li explained in her interview, “I forgot the part of speech for ‘basic’. I revised it.”

The think-aloud data and interviews indicated high motivational strategy use by the four cases when revising based on teacher feedback. Despite its lower frequency, teacher feedback was all related to meaning, demanding that learners understand the feedback and revise accordingly. The four cases incorporated every item of teacher feedback. Think-aloud data indicated that all meaning-level teacher feedback caused learners to self-reflect and evaluate their original drafts. This process, as the interviews also showed, is characterized by high motivation and constant emotional control.

Notably, learners employed several strategies to achieve their feedback revision goals. The highest and lowest proficiency learners, Luan and Li, are cited here as examples of a series of SRL strategies used when revising based on teacher feedback. Both Luan (high-level) and Li (low-level) employed a series of metacognitive and cognitive strategies when making revising decisions. They both reflected on feedback, evaluated it with reference to their original essays, interpreted the reasons for their inappropriate use, and revised accordingly. They also turned to others for help (social strategies), especially when they were unsure about some indirect teacher feedback. This shows that motivational strategies are vital during this process; they help learners regulate their own motivation to keep revising in a positive frame of mind. Subsequently, they can orchestrate the array of strategies at their disposal to revise effectively.