The Effects of Coded Focused and Unfocused Corrective Feedback on ESL Student Writing Accuracy

Purpose: This study adopted a mixed-method approach, including a classroom experiment and 24 in-depth interviews, to investigate the effects of two feedback techniques (coded focused and unfocused written corrective feedback) on ESL learners’ writing in a self-financed tertiary institution in Hong Kong. Methodology: Three intact classes of 47 students served as the experimental and control groups; the control group only received feedback on content and organization, whereas the two experimental groups also received focused and unfocused linguistic feedback, respectively. The feedback intervention was conducted over an eight-week intensive summer course, focusing on three grammar errors (articles, singular/plural nouns and verb forms). Altogether, students wrote seven pieces, four of which were analysed for the present research. Results: The study found that students who received focused written corrective feedback (WCF) significantly outperformed the other two groups, though the effects varied across error types. Meanwhile, no significant differences were found between the unfocused and control groups. In-depth interviews explored how individual learners’ metalinguistic understanding and engagement affect their intake of WCF. The results revealed that learners who received focused feedback developed a deeper understanding of the linguistic nature of specific error types. Learners with limited English proficiency were less likely to apply their linguistic knowledge to revise a task or write a new one. Conclusion: Because not all errors deserve equal attention, teachers and students should consider how feedback can be used more effectively, particularly in areas where comprehensive feedback is considered obligatory. When teaching students with limited language proficiency, it is recommended that, rather than providing a wide range of error corrections, teachers provide focused feedback complemented with carefully designed metalinguistic support.


INTRODUCTION
Written corrective feedback (WCF), also referred to as grammar correction or written error correction (Ferris et al., 2013), involves "correction of grammatical errors for the purpose of improving a student's ability to write accurately" (Truscott, 1996, p. 32). WCF has long been widely utilised and recognised as an integral part of feedback in L2 writing across different educational levels and institutions around the world. Given its essen-tial role in L2 writing instruction, WCF is a topic that has been brought up repeatedly over the past four decades.
As far back as the 1980s and early 1990s, empirical studies were being conducted to examine the effects of WCF on English as a Second Language (ESL) students (e.g. Fathman & Whalley, 1990); English as a Foreign Language (EFL) students (e.g. Robb et al., 1986); and students of other foreign languages (e.g. Lalande, 1982;Semke, 1984).The results were Since then, many more studies have been conducted on the efficacy of WCF. These studies often examined and compared the effectiveness of several types of WCF, such as focused and unfocused WCF and direct and indirect WCF. Their findings not only demonstrated whether WCF is effective but also which type of WCF is more effective. Although these studies, which mainly adopted quasi-experimental designs, have contributed a great deal of valuable knowledge and insight about the efficacy of WCF in L2 writing, there are still many under-research topics. For example, there is scant research directly comparing the effectiveness of focused and unfocused WCF (Mao & Lee, 2020). Moreover, there are methodological gaps in this research field. As many researchers have correctly pointed out, most of the existing empirical studies of WCF were either experimental or quasi-experimental studies conducted in controlled research environments. It is, therefore, doubtful whether or not-and if so, to what degree, these studies' findings are applicable to L2 writing instruction in real classrooms (Ferris et al., 2013;Lee, 2013;Mao & Lee, 2020;Storch, 2018). To respond to the recent call "for stronger ecological validity in WCF research" (Mao & Lee, 2020, p. 10), and to fill the two aforementioned research gaps, the present study adopted a mix-method approach to investigate the relative effectiveness of the focused and unfocused WCF and to determine how learners' engagement with WCF affects the intake of teacher feedback in EFL classrooms at a college in Hong Kong.

Effectiveness of WCF
Whether or not WCF is effective appears to be an old question, but the findings in the literature remain inconclusive. First, there has been abundant evidence in favour of WCF (Bitchener et al., 2005;Ferris & Roberts, 2001;Lalande, 1982). Kang and Han (2015) conducted an influential meta-analysis of 21 experimental or quasi-experimental studies on WCF in L2 from 1980 to 2013 and found that WCF could increase L2 writing learners' grammatical accuracy with a "small to moderate" (p. 10) overall effect size. More recent research has also demonstrated the benefits of WCF (e.g. Frear & Chiu, 2015). However, some studies have had less positive findings. For example, in a study of 80 intermediate-level ESL students at a US college, Sheen et al. (2009) found that corrective feedback was not more effective than mere writing practice in increasing the students' linguistic accuracy. Moreover, in another study on ESL learners at an American university, Evans et al. (2011) found that the group that received detailed corrective feedback in the traditional process writing approach even had poorer accuracy over time. Given the inconsistent evidence concerning this essential matter, it is necessary to do more research examining whether WCF is effective in helping L2 writing learners improve their linguistic accuracy.

Effectiveness of Different Types of WCF
The hope to find more effective ways of providing WCF has led to more inquiries into the relative effectiveness of different types of WCF. Studies are often conducted on WCF of different scopes, namely, focused vs unfocused WCF. Focused WCF refers to "correction that is provided for specific error types" (Ferris et al., 2013, p. 309); by comparison, unfocused feedback "lacks a focus" (Lee, 2017, p. 169). In addition to the scope of feedback, the effects of the explicitness of WCF have been examined through comparisons of direct and indirect WCF (Xie & Lei, 2019). When teachers provide direct WCF, they not only identify the error (usually by underlining or circling it) but also provide the correct form above or beside the error (Bitchener et al., 2005;Ferris et al., 2013;Rahimi, 2021). On the other hand, indirect WCF merely indicates the existence of an error but does not provide the correct form (Ferris et al., 2013;Mao & Lee, 2020); thus, students are left to correct their errors by themselves (Bitchener & Knoch, 2008). An alternative feedback instrument that is frequently used in combination with WCF is metalinguistic feedback, which "offers learners metalinguistic explanation and examples through error codes and correct usage" (Mao & Lee, 2020, p. 6). Some researchers believe that metalinguistic feedback can develop L2 writers' awareness and elicit their explicit grammatical knowledge (Mao & Lee, 2020). Ferris et al. (2013) advocated that explicit corrective feedback like metalinguistic explanation may be especially advantageous to EFL learners who have learned a significant amount of formal grammar, "as the codes, corrections, or explanations may elicit their prior knowledge" (p. 309). Mao and Lee (2020) reviewed 59 empirical studies related to WCF scope published in high-impact journals from 1979 to 2018. They found that previous research has mostly examined either comprehensive or focused WCF separately, and that these studies' findings are mixed and inconclusive. They also found that by 2018 only three studies (Ellis et al., 2008;Frear & Chiu, 2015;Sheen et al., 2009) had directly compared the effects of focused and less focused/comprehensive WCF, highlighting a strong need for comparative studies on focused and unfocused WCF. Recently, Rahimi (2021) has also examined the relative effectiveness of focused and unfocused WCF. The four studies which directly | Research Article compared the efficacy of focused and unfocused WCF are reviewed individually below. Sheen et al. (2009) conducted a nine-week quasi-experimental study at a US college, dividing 80 intermediate-level ESL students into four groups: a direct focused WCF group, a direct unfocused WCF group, a writing practice group and a control group. All three treatment groups were required to complete two written narrative tasks, whereas the control group was not. The direct focused group was corrected on errors in the use of indefinite and definite articles; the direct unfocused group was corrected on five error categories (articles, the copula 'be', regular past tense, irregular past tense and prepositions); and the writing practice group received no feedback. It is worth noting that the writing practice group was the study's actual control group, whereas the group labelled "control" could not be validly compared to the WCF group, as it was not even assigned the writing tasks. The results suggested that focused WCF is more effective than unfocused WCF in improving the accurate use of articles in both the short and long term. A more striking finding was that the unfocused group did not even outperform the "control" group, which neither performed the writing tasks nor received correction; in other words, this study "failed to demonstrate any benefit in providing unfocused CF" (Sheen et al., 2009, p. 567). The authors concluded that focused WCF is beneficial because it can help learners (1) spot the mistakes in their composed work, (2) systematically conduct hypothesis testing and (3) use their explicit grammatical knowledge to monitor the linguistic accuracy of their writing. By contrast, unfocused WCF may be "a confusing, inconsistent and unsystematic way" of providing corrective feedback; moreover, it may overwhelm learners (Sheen et al., 2009, p. 567). Ellis et al. (2008) conducted a ten-week quasi-experimental study on 49 intermediate-level EFL students at a Japanese university. The students were divided into three groups, all of which completed three picture narrative tasks at different stages of the research. After writing each task, the focused WCF group was directly corrected only on definite and indefinite articles, the unfocused WCF was directly corrected on various grammatical errors, including article misuse, and the control group was not corrected at all. No revision was required. All three groups also took a pretest, an immediate posttest (on the same day of receiving WCF), and a delayed posttest (four weeks later). The study showed that both treatment groups made more notable progress than the control group in terms of grammatical accuracy. Although there seemed to be no significant difference between the performance of the focused group and that of the unfocused group, the study indicated that focused WCF "may be more effective in the long run" (Ellis et al., 2008, p. 367). Frear and Chiu (2015) conducted a three-week quasi-experimental study on Chinese EFL learners at a university in Taiwan, examining the relative effectiveness of focused and unfocused WCF on the accuracy of weak verbs (regular verbs) and the total accuracy of all structures in new pieces of writing. They divided 42 students into three groups: a focused indirect WCF group, an unfocused indirect WCF group and a control group. The results showed that both focused and unfocused WCF groups outperformed the control group not only in the immediate posttest but also in the delayed posttest; however, there were no differences in the efficacy of the two types of WCF.
In a recent study, Rahimi (2021) assigned 78 French-speaking Canadian learners of intermediate-level ESL into four groups: comprehensive revision, comprehensive non-revision, focused revision and focused non-revision. The results showed that focused WCF was more effective than comprehensive WCF in facilitating the learners' reduction of the targeted errors, especially those which were "more complex and more cognitively difficult to process" (Rahimi, 2021, p. 704).
In short, there is a general belief that, compared to unfocused WCF, focused WCF can reduce writing learners' attentional and cognitive burdens such that the learners are more likely to notice their mistakes (Frear & Chiu, 2015;Mao & Lee, 2020). While some researchers advocate focused feedback, others point out that comprehensive WCF may have greater ecological validity than highly focused WCF in real classrooms (Bruton, 2010;Ferris, 2010;van Beuningen, 2010). As there are few studies comparing the effectiveness of focused and unfocused WCF, the existing findings about the relative effectiveness of the two types of WCF are inconclusive. Moreover, there is even scarcer research on the use of the two types of WCF together with metalinguistic feedback. Therefore, the current study aims to fill this gap by comparing the efficacy of focused and unfocused WCF used in combination with a metalinguistic form.

L2 Writing Learners' Engagement with WCF
As Han and Hyland (2015) pointed out, "learner engagement is a critical link that connects the provision of WCF with learning outcomes" (p. 31). Engagement refers to "how learners respond to the feedback they receive" (Ellis, 2010, p. 345). It may be affected by the type of corrective feedback, learners' individual differences and contextual factors, and it can be investigated from cognitive, behavioural and attitudinal perspectives (Ellis, 2010). Based on Ellis's (2010) framework for engagement with corrective feedback, Han and Hyland (2015) conducted a case study involving four EFL learners at a Chinese university. They found that learners might disregard WCF or mistake it as content feedback or praise. Their findings also showed that noticing errors does not equal understanding, and that EFL learners tend to process WCF "at the surface level" or even use "avoidance strategies" (p. 40). Furthermore, during the revision process, all four students carried out both cognitive strategies (e.g. retrieval of prior knowledge, memorization, and conceptualisation of details) | Research Article and meta-cognitive strategies (e.g. making plans, setting priorities, evaluating, and monitoring); however, the effectiveness of these strategies was mainly determined by the learners' individual characteristics, such as "learning goals, beliefs about the effectiveness of WCF, about English writing, and about their own writing abilities" (p. 40). Regarding behavioural engagement, all four students consulted external resources and could draw on them to correct errors even if they did not clearly understand metalinguistic rules. This suggests that the use of external resources alone is not concrete proof of extensive engagement, whereas "the quality of external resources and the way these resources are used" (p.40) make a major difference. With respect to affective engagement, on one hand, the findings seemed to support Truscott's (1996) point that WCF may arouse negative emotions that hinder L2 learners' learning of writing; on the other hand, there is evidence that learners' affective responses may be influenced by their expectations and self-beliefs, and that negative feelings may be controlled or even converted into something positive such as motivation. In short, Han and Hyland's (2015) study demonstrates that learner engagement with WCF involves a complex interplay of various factors on multiple dimensions.
A few other studies have also investigated students' engagement with WCF from various perspectives. Hyland (2003) conducted a case study involving six ESL students at a university in New Zealand to explore their beliefs about and attitudes towards WCF. The students enthusiastically welcomed teachers' WCF and incorporated the feedback into subsequent revisions to varying degrees. All six students felt that such feedback may not have immediate effects, but that "repeated feedback would eventually help them, and that without the feedback they would fail to note the errors and improve" (Hyland, 2003, p. 228). Chen et al. (2016) investigated learners' perceptions of WCF and their WCF preferences through open-ended survey questions posed to 64 EFL students at a Chinese university. These students also reported that they valued WCF and considered it an important tool for learning. However, they expressed dissatisfaction with the lack of interaction in WCF. The desire for more interactive corrective feedback has also been found in other studies. For example, a 16-week multiple-case study on ESL students at a US university highlighted the necessity of providing opportunities for students to discuss their WCF with their teachers (Ferris et al., 2013).
In a large research project on the efficacy of direct and indirect comprehensive WCF for ESL students (predominantly Chinese graduate students) in Australia, Storch and Wigglesworth (2010) conducted a case study exploring individual students' engagement with corrective feedback in revision and new writing. The findings showed a higher level of student engagement with indirect WCF than with direct WCF. The results also indicated a link between the nature of the errors and students' uptake and retention of the WCF. Specifically, for feedback on superficial errors such as me-chanics, uptake and retention can occur with "limited or no overt engagement" (p. 328). The study also suggested that "learners' beliefs, attitudes toward the form of feedback received, and their goals seemed to have an effect on whether the feedback was retained" in new writing (p. 328).
To summarize, it is necessary to conduct more carefully designed research on WCF, particularly on topics such as the relative efficacy of focused and unfocused WCF (Mao & Lee, 2020). As Bruton (2009) has argued, the research design must fulfil certain basic requirements, such as including a control group without WCF and a post-test with a new writing task. A control group is essential to ensure that students' grammatical performance can be attributed to the feedback condition and not to other factors (Truscott, 1996). Moreover, most of the existing empirical studies have employed a quasi-experimental design; for the findings to have greater pedagogical value, the efficacy of WCF must be investigated in more ecologically valid contexts such as authentic writing tasks and real, intact classrooms (Evans et al., 2011;Mao & Lee, 2020). Additionally, it is necessary to include qualitative methods in the research to obtain a more thorough and comprehensive understanding of the effects of WCF. Therefore, the current study adopted a mixed-method approach to address the following questions in three authentic EFL classrooms at a college in Hong Kong: RQ1: Is written corrective feedback useful in helping EFL students improve their linguistic accuracy in subsequent revision and new writing?

RQ2:
While the other factors remain constant, which type of WCF, focused or unfocused, is more effective in helping EFL students improve their linguistic accuracy in subsequent revision and new writing?
RQ3: How does EFL students' engagement with WCF affect their linguistic accuracy in subsequent revision and new writing?

METHODS
The present study was designed to respond to the gap of a mixed-method approach in the WCF research. The quantitative segment investigated the effects of two different feedback treatments on groups of learners in an authentic classroom setting during an 8-week semester (RQ1 and RQ2). It involves 47 students' writing collected from three intact classes. We also use interview data to explore how learners' engagement with teacher feedback affect the effect of WCF (RQ3).

Pedagogical Context
The study involved 47 first-year ESL learners (46.7% males and 53.3% females) from two bachelor's degree pro-

| Research Article
grammes at a self-financed college in Hong Kong. More than 60% of these students had scores of Level 2 or below out of five levels (equivalent to IELTS 5.0 or below) on the English subject of the Hong Kong Diploma of Secondary Education Examination, which is the university matriculation examination in Hong Kong. As such, they did not meet the minimum requirements for government-funded bachelor's degree programmes. Thus, the English language proficiency of the participants was located at the weaker end among Hong Kong degree students.
The students were enrolled in an intensive 8-week summer course on English professional communication that included 39 classroom teaching hours over 13 lessons, "the same number of teaching hours as in a regular 13-week semester.
The intervention was conducted during the second half of the course, which lasted four weeks (6 lessons). The course was taught by three experienced teacher-researchers, who were all familiar with the course content, expected learning outcomes, assessment requirements and marking rubrics.

Classroom-Experimental Design
The study used a classroom-experimental design involving three intact classes that served as two treatment groups, focused WCF (N = 15) and unfocused WCF (N = 18), and a control group (N = 14). The pretest result indicated no significant differences among the three groups from the outset of the study. All the students were taught with the same materials and were assessed on the same tasks. The intervention had all three groups writing the same set of tasks but receiving different types of feedback.
The focused group received feedback on the three most prominent linguistic error types identified by the marking of the pretest task: articles, singular/plural nouns, and verb forms. Prominent, here, means "most frequent" or "most serious" for text effectiveness. The unfocused group received feedback on 15 types of linguistic errors including the three chosen for the focused group. The additional 12 linguistic error types included word choice, run-ons, pronouns, upper or lower cases, word forms, voice, verb tenses, prepositions, comparative or superlative forms of adjectives, the verb be, subject and verb agreement and fragments (Appendix 1). The control group received no grammar-corrective feedback. To satisfy ethical requirements, all three groups were given feedback on the quality of content and organisation. This is an advantage of our design because it reflects the genuine feedback students would normally receive in natural classroom settings.
There were six treatment sessions (Table 1). In classes 8, 10 and 12, students were asked to write a task in class. They received the feedback in the next lesson (classes 9, 11 and 13), and were then asked to use the feedback to rewrite the task. All three groups were reminded to improve content, organization and language accuracy when rewriting a task.
To provide further guidance for the rewriting activities, the focused group was provided with a form containing meta-linguistic explanations of the three chosen grammatical items (Appendix 2), and the unfocused group was provided with a meta-linguistic explanation form containing 15 grammatical items (Appendix 1). The marking codes, their meaning, and examples were provided in the metalinguistic form. For example, Code 5 refers to the common mistake known as run-ons. The meaning of this code is displayed in the next column: "Run-on sentences include (1) fused sentences (no punctuation at all between two independent sentences) and (2) comma splices (two independent sentences are divided by a comma)". Two erroneous sentences are provided as examples: "I like listening he likes talking"; "I like listening, he likes talking". Four correct sentences are provided in the next column to demonstrate how this type of error can be revised (e.g. "I like listening, but he likes talking.").
All in-class professional writing tasks required the use of various linguistic forms to maintain an appropriate tone, achieve coherence and adopt an appropriate style. The stu- | Research Article dents were given 50 minutes to write each task in class. Because all the in-class tasks were similar to the task in the final examination, the intervention was naturally integrated into classroom teaching. We used four of seven writing samples to measure students' performance. Task 1 served as the pretest, while Revision 1 was the subsequent revision, Task 2 was the immediate posttest (3 days after a lesson) and Task 4 was the delayed posttest (examination, four weeks after the first task) ( Table 1). The two posttests explore whether students can benefit from WCF in different writing conditions, with Task 2 being open-book, no time constraint and low stake and Task 4 being closed-book, timed and high stake.

Quantitative Data
To ensure marking reliability, two random samples from each group were marked and discussed by the researchers, following which a new sample was marked separately to ensure marking consistency. After achieving over 90% agreement, the rest of the scripts were distributed to individual raters who had been keeping close communication during the marking process. After the marking was completed, 40% of the scripts were selected randomly from each group for a second rater to re-mark. Intra-class correlation coefficients were calculated (via SPSS: two-way mixed model; absolute agreement; single measure) to estimate inter-rater reliability for each of the three error types. The inter-rater relia-bility indices were found to be .969 for verb forms, .997 for articles, and .940 for singular/plural nouns, indicating that the raters had achieved an excellent level of agreement in coding the errors (Koo & Li, 2016).
Error scores were calculated by normalized error counts, a method suggested by Ferris and Robert (2001). This procedure consists of dividing the number of errors by the number of words in the writing sample and then multiplying it by a standard number representing the average number of words of all writing samples. In this case, the standard number used was 320, i.e. the average length of the samples. The normalized error counts enabled us to compare the numbers generated from different writing tasks across different groups at different times.
The error scores were then compared across groups and over time to answer RQ1 and RQ2. Due to the relatively small sample sizes of the experimental and control groups (N = 15, 18, 14) and the fact that the variables tended to have skewed distributions, a more general non-parametric test (Kruskall-Wallis One-Way ANOVA) was conducted to compare the error statistics across groups. As its name suggests, the Kruskall-Wallis test is conceptually equivalent to its parametric ANOVA test, but it is more general and accommodative and does not have assumptions about data normality. Similarly to ANOVA, the Kruskall-Wallis test starts with an overall test to determine whether there is a  In the second post-hoc step, pair-wise comparisons are conducted among all possible pairs. To correct the inflated alpha level due to multiple comparisons, Bonferroni correction was applied to produce a much stricter criterion of statistical significance (adjusted p.).

Individual Interviews
To aid the interpretation of our quantitative data, 24 oneon-one interviews ( Table 2) were conducted with six student participants, comprising three post-revision retrospective interviews (Appendix 3) and a reflective exit interview (Appendix 4). All the interviews took about 50 minutes and were video recorded. The six participants (Table 3), three from each treatment group, were chosen by the teacher-researchers after the first revision task was completed. The criteria for choosing an interviewee included the willingness to submit drafts, different levels of language proficiency and availability for participation. Retrospective interviews, carried out within three days of a specific revision lesson, aimed to investigate (1) the extent to which learners understood the WCF, and (2) how learners engaged cognitively with the WCF in their writing. The reflective exit interview was conducted after the examination (T4) at the end of the semester. The interview questions covered various topics, including students' general experience with WCF and the strategies they used for engaging with WCF throughout the course.

Interview Data Analysis
The interview data were analysed according to the principles of inductive analysis (Merriam & Tisdell, 2015). Following the full transcription of the interviews, all the raw data were repeatedly studied and thematically coded (e.g. understanding of WCF, WCF engagement strategies and writing conditions) according to the research questions. The researchers worked together to synthesize ideas, combine similar codes, resolve differences and condense the data into more specific codes pertaining to how students perceive and use WCF.
The specific codes include, for example, "avoidance as a strategy", "superficial or deep level of learning" and "use of WCL in different conditions". To enhance the trustworthiness of the findings, different data sets were triangulated, including the students' responses to questions and their actual applications of WCF in subsequent writing.

RESULTS
Results are presented, herein, to address each research question in turn, beginning with the quantitative results comparing the three groups' error codes over time (RQ1 and RQ2), and followed by the interview results illuminating how individual learners perceive, respond to and apply WCF in their writing (RQ3).
Kruskall-Wallis tests were applied to four error scores (namely, the overall score and the scores of three error types) to compare the control, focused and unfocused groups across four writing samples (T1, R1, T2, T4). When the overall test was significant, post-hoc, pair-wise comparisons were conducted to identify which two groups were significantly different. Bonferroni adjustment was applied to the significance level of post-hoc comparisons, and generated stricter adjusted significance statistics. No significant difference was found among the three groups in the pre-test (T1) in the overall scores or the three individual error scores, which means the three groups had comparable performance before the intervention. After the invention, however, significant differences were found among the three writing samples (R1, T2 and T4) in terms of one or more of the error types.

Overall Effectiveness of WCF (RQ1) and Effectiveness of Focused vs. Unfocused WCF (RQ2)
The performance of the control group, which did not receive WCF, was compared with that of two experimental groups, one of which received focused feedback and the other unfocused feedback. Due to space limit, Table 4 only presents the results that are found to be statistically significant; for the full results of the analysis, interested readers can refer to Appendix 5. Figure 1 shows a comparison of the average numbers of overall errors among the three groups. Figure 2 shows a comparison of the average error numbers of each error type among the three groups.
First, no significant difference was found between the control group and the unfocused group (C-UF) in any of the three error types across all four writing samples. This means that the unfocused group produced as many errors as the control group on linguistic accuracy in both the subsequent revision and the new task.
On the other hand, the focused group performed significantly better than both the control group and the unfocused group. It outperformed the control group (F-C) in terms of overall performance (test statistics: -3.245, Adj.p=.004) and singular/plural nouns in R1 (Adj.p=.022). It also outperformed the unfocused group (F-UF) in terms of overall errors in R1 (Adj.p=.011) and articles in R1 (Adj.p=.004) and T2 (Adj.p=.011).
However, no significant difference was found in R1 between the focused group and the control group (F-C) regarding the use of articles and verb forms. Similarly, in the two new tasks (T2 and T4), the differences between the two groups were not significant. Between the focused group and the unfocused group (F-UF), the differences in the use of singular/ plural nouns or verb forms were not statistically significant. Table 4 shows the advantages of focused feedback. The focused group performed significantly better than the unfocused group in its use of articles in the subsequent revision (R1) and in one of the new tasks (T2); it also performed better on verb forms in one of the new tasks (T4), though not to the point of statistical significance (T4, p=.068).

| Research Article
Compared to the unfocused group, which had to deal with up to 15 different error types in WCF, the focused group only had to attend to three different error types, at most, so these students probably had more cognitive space for internalizing the WCF and correcting the errors.
Taken together, the findings for RQ1 and RQ2 show that WCF is more effective than self-revision in helping EFL students improve their linguistic accuracy in subsequent revision and new writing (RQ1), but only focused WCF shows significant effects for the participants of this study (RQ2).
A closer examination of when the focused feedback showed effects (in R1, T2, or T4) revealed interesting differences among the three error types. Below is an interpretation of why error types may influence the effects of focused WCF.
Errors in the use of singular/plural nouns and articles can be corrected mechanically without a sufficient understanding of their usage as long as the teacher provides an error code. Errors in verb forms, however, cannot be corrected easily, because choosing the correct verb form among several requires a deep understanding of the usage of verb forms in different sentence structures. The following are some verb form errors that the students made in R1 and T2. In the sentence "We should try to change our billboards to some crowded areas like MTR stations", a student wrongly used "try to change" to indicate a tentative suggestion. Choosing the correct verb form, in this case, requires knowledge of the meanings of "try to do" and "try doing". In the sentence "The table shows that there are nearly half of the teachers choose i-class for e-learning", the student failed to use the -ing form "choosing" as a postmodifier of the noun phrase "the teachers". Selecting the correct verb form, in this case, requires applying the English sentence structure rule of only one main verb per clause. While some students made the mistake of including two base-form verbs in one clause, some others wrote a clause without a main verb in the base form; for example, "Radio is not a cost-effective method of advertising, which accounting for 31% of the advertising expenditure". These three examples show that a simple error code, i.e. "verb form", may not be enough for students to choose the correct form. They need to know how particular verbs take different forms, i.e. the infinitive or gerund.
They also need to have good knowledge of English sentence structures in order to use the correct verb form in different structures. The first time the students were required to revise their first drafts in R1, the focused WCF seemed to be immediately effective for errors that could be corrected mechanically. The focused group performed significantly better than the control group in the use of singular/plural nouns and the unfocused group in the use of articles. However, the effect of focused feedback on verb forms, a construction Furthermore, the linguistic accuracy in the use of singular/plural nouns or articles that the focused group gained in R1 may not have been carried on to new tasks (T2 and T4). The only exception is that the focused group performed significantly better than the unfocused group on articles in T2, a new task that immediately followed R1; however, the improvement was not maintained in the delayed posttest (T4). Interestingly, with regard to verb forms, although the focused group did not perform significantly better than the unfocused group in the subsequent revision, its performance was better than the unfocused group in T4 (p=.023, adj. p=.068). It is possible that some participants in the fo-cused group corrected their errors in singular/plural nouns or articles mechanically, based on error codes. Without a thorough understanding of usage, their errors were not reduced in the new tasks. It seems that the linguistic accuracy gained by mechanical correction cannot be carried on to new tasks. As to verb forms, a grammatical construction that students may not be able to correct mechanically, the participants might have had difficulties choosing the correct verb form in the subsequent revision based on simple error codes. Deeper learning of this grammatical construction might have taken place in the focused group, as they engaged in problem-solving in the tasks later in the intervention, namely in T2, R2, T3 and R3. It is not surprising that this group's improvement was observed in the delayed posttest (T4) but not in T2, as deeper learning takes time.

Figure 1
Comparison of overall errors among groups

Figure 2
Comparison of errors of each type among groups

Effect of Students' Engagement with WCF (RQ3)
This section uses interview data to explain why focused WCF was more effective than unfocused WCF and how learner engagement might have influenced the effects of the two types of WCF. Three major themes emerged from a comparison of the two groups' interview transcripts: (1) engagement strategies, (2) level of learning (deep versus superficial learning) and (3) writing in different conditions.

Engagement Strategies
A specific theme emerging from the data was how students' level of metalinguistic understanding affects the engagement strategies learners employ. Three engagement strategies emerged from this study's data: (1) avoidance, (2) mechanical correction and (3) correction through the application of grammatical rules.

Avoidance
One strategy the students used was avoidance, i.e. avoiding difficult words or grammatical structures. When Tang, a student from the unfocused group who demonstrated a low level of metalinguistic understanding, was asked to read the definitions and examples in the metalinguistic form (Appendix 1), she struggled with the definitions because she could not understand the difficult words or technical terms. When comparing the two drafts of Tang's first task, it was found that she only revised the errors in the first two paragraphs. Tang explained why she did not revise the second part of her writing.

Mechanical Correction
Some students demonstrated a partial understanding of the WCF. Below is the example of Li (unfocused group). In the first interview, after reading its definition, Li explained the code "fragment" by engaging with the example.
Li: I think fragment means something is missing.
Researcher: Please read the example (on the error code sheet) and explain which part is the fragment.
Li: Since I missed last class. I did not know the homework. Since I missed last class, I did not know the homework.

Researcher: (Pointing to the sentences in the example). Which one is the fragment?
Li: Both are fragments. The meaning is incomplete when one stands alone without the other.
Li's performance can be interpreted as having developed a partial understanding of this code, i.e. that "fragment" means "something is missing". However, her explanation shows that she only examined whether the meaning was complete and did not successfully employ knowledge of English syntax. Without thorough understanding, Li could not apply the target linguistic form to solve her problem. Below is the comparison of Li's two written drafts: Task 1: Radio is not a cost-effective method of advertising, which accounts for 31% of advertising expenditure. But it just 17% of customers <fragment>.
Revision 1: Radio is not a cost-effective method of advertising, which accounts account for 31% of advertising expenditure, but it just 17% of customers <fragment>.
Li explained how she applied the code to solve her problem: "I saw the fragment code and realized that something is missing. The example shows that I can correct it by putting the two parts together". Here, Li can be interpreted as having overgeneralized how to correct a fragment due to a lack of deep understanding of syntax. It appears that the metalinguistic definitions and example(s) may not have been sufficient for some participants to understand and apply grammar rules.

Applying Grammar Rules in Writing
In the exit interview, Leung (focused group) was asked to correct the grammatical error related to the use of verb forms in the sentence "Read romance novels is relaxing". Not only did Leung solve the problem, but he also clearly explained how to apply the rule regarding verb forms: Leung: "Read" should be changed to "reading" because we cannot use a base form verb as the subject of a sentence.
Leung further explained how he learned this grammatical structure:

Summary
The findings show a connection between learners' metalinguistic knowledge and their manner of engaging with teachers' feedback. It seems that the more metalinguistic knowledge learners acquire, the more likely they are to apply grammatical rules in their writing. Focused WCF has been observed by this study to be more beneficial due to its more manageable cognitive load, particularly for students with moderate language proficiency.

Deep Versus Superficial Learning
A careful comparison of the interview data for students from the focused and unfocused groups shows that the focused group demonstrated deeper learning than the unfocused group. When asked how they engaged with WCF, all three participants from the focused group mentioned explicitly and repeatedly that they advanced their learning of a specific grammatical structure through additional sources and by various methods. All three participants from the focused group tried to use additional methods to deepen their understanding of the target linguistic forms, for example by using a portfolio to analyse error patterns, seeking help from peers, checking grammar books, exploiting their grammar knowledge rather than depending on grammar-checking tools, and reflecting on their experiences. All of these strategies worked together to enhance the usefulness of WCF.
By contrast, none of the participants from the unfocused group mentioned specific grammar items when they shared their experiences of WCF; rather, all vaguely mentioned that they used "the codes" or "the metalinguistic form", indicating that they were not deeply engaged in the learning of specific grammatical features.
For example, Li and Tang (unfocused group) did not demonstrate effort other than to check the metalinguistic form.
Chan was the only student from the unfocused group who used some additional sources to further her study. However, in the exit interview, she could not explain in depth how to solve problems, as shown in the following example: In fact, I do not know exactly when to use a specific verb form.
When the teacher pointed out that there was a verb form prob- Tang and Li struggled with some of the codes until the end of the semester. Chan tried to learn more deeply by using extra resources, but it appeared to be difficult for her to manage so many grammar items during such a short period of time.

Summary
The three interviewees from the focused group all demonstrated deeper learning than those from the unfocused group. This can be interpreted to mean that learners from the focused group were more likely to pay attention to feedback directed at a limited number of linguistic error types, while their counterparts may have been cognitively overloaded by a wide range of linguistic problems.

Engagement in Different Writing Conditions: Revision Versus New Tasks
This section examines whether students from different groups interacted differently with teacher feedback in different writing conditions. Both groups reported that they paid more attention to grammar in revision, while they had to maintain a balance between language accuracy and content in a new task. Table 5 summarizes the number of errors found in the pretest (T1) that students were required to revise in Class 9 (R1). The revision was obviously much more manageable for the | Research Article students from the focused group, who had significantly fewer errors and error types. By contrast, the three interviewees from the unfocused group had to work with more than 35 errors representing more than 10 error types.

Revision
In the first retrospective interview, when asked about his experience with the revision task, Ma said the following: It seems that the focused group (of which Ma was a representative) was in a privileged position for using the teacher's feedback because the group had "a clear direction to work on". By contrast, the workload was much heavier for the unfocused group.

New tasks.
In retrospective interview 2, when asked how they worked with a new writing task (T2), Leung and Tang both referred to their efforts at balancing different assessment components in writing a new task. While Leung could manage a good balance, Tang reported difficulties in doing so. Similar findings emerged about writing a new task in the examination. All students (from both groups) reported a certain degree of anxiety about the closed-book examination. But, due to their heavier workload, the unfocused group's anxiety was more pronounced:

Summary.
Focused WCF was found consistently to benefit the learning process in all three writing conditions (R1, T2 and T4); the more manageable workload it entailed allowed learners to direct their focus towards revising grammar in a revision task or allocating efforts to different components of a new task.

DISCUSSION
This section begins by answering a fundamental question regarding whether WCF is beneficial to the accuracy of L2 writing (RQ1). The quantitative data analyses found that students who received coded focused feedback produced significantly fewer errors than students in the other two groups. Meanwhile, no significant differences were observed between the unfocused and control groups across all four writing samples. Certain types of WCF, then (coded focused WCF, but not unfocused WCF, in this case), can be considered conducive to the improvement of L2 writing accuracy. In other words, Ferris's (1999) argument that effective WCF can help at least some student writers, providing that the right WCF is managed properly, is supported.
Regarding the question of whether focused or unfocused WCF is more beneficial to L2 writers (RQ2), feedback scope and error types were both found to potentially play a role.
In contrast to Lalande's influential 1982 study, which reported success using a large number (12) of error categories, the present study's findings suggest that L2 learners with limited language proficiency can only cope with a few (3) error categories at a time. While it is clear that the focused group outperformed the unfocused group, the question of which error types are more effectively addressed in which writing conditions (redrafting, new writing or both) seems to be a complex one. Article errors were significantly reduced by the focused group in revision (R1, Adj.p=.004) and open-book in-class writing (T2, Adj.p=.011), but not in a more stressed condition (T4). By contrast, contrary to Ferris and Roberts' (2001) findings about the significant improvement of verb forms in revision, students in the present study did not seem to be able to reduce the error counts of verb forms in the first two writing attempts (R1, T2). The error counts were reduced by the end of the semester, though not to the point of statistical significance (T4, Adj.p=.068). Interestingly, despite greater pressure in the delayed post-test, which was a high-stake examination, the students could still use the targeted grammatical item more accurately, which may strongly demonstrate the efficacy of indirect focused WCF.
On the other hand, the fact that the students did not show progress in using these items in the revision and first new | Research Article writing tasks may suggest the complexity of learners' development of linguistic skills (Benson & DeKeyser, 2019). The present study reinforces Mao and Lee's (2020) findings that the nature of errors can affect learners' uptake of feedback.
The qualitative data herein identifies two salient factors that could facilitate or constrain the effect of WCF (RQ3): (1) the level of metalinguistic understanding and (2) the ways in which learners engage with WCF. The revision of a singular/ plural noun error is obvious because students can simply use the other form when the marked form was incorrect. Conversely, the grammatical rule acquisition of verb forms appears to require more time and effort. The ability to apply a grammatical rule in a new piece of writing requires an even more thorough metalinguistic understanding. In the final examination, many participants in this study mistakenly used the plural noun "users" in the phrase "user forecast", where "user" is a modifier and not a typical plural noun. In order for deep metalinguistic understanding to occur, learners' attention and engagement must be directed to specific grammar features. This kind of deep learning was observed in the focused group (Leung, Ng and Ma), where learners could interact intensively with a limited number of specific grammatical forms, but not in the unfocused group (Chan, Li and Tang).
This resonates with Sheen et al. (2009) and Bitchener and Knock (2008) in that learners who received focused feedback developed a deeper understanding of the linguistic nature of specific error types. Cognitive theories of L2 acquisition (Gass, 1997;Schmidt, 2001) have also provided solid theoretical reasons for focused WCF by establishing a strong link between understanding and L2 development.
Learners' language proficiency is another main factor which effects students' intake of WCF. The theoretical assumption is that focused WCF is likely to benefit learners with limited capacities, as it imposes a less heavy cognitive load and thus provides more scaffolding for grammar acquisition (Van Pattern, 2004). Our results echo the above cited literature in that although the students from both treatment groups appeared to have raised linguistic awareness, the unfocused group did not demonstrate deeper learning of new linguistic features due to the constraints of the heavy cognitive load brought about by the large number of errors in their writing.
All participants in the present study reported that metalinguistic information increased their awareness of the targeted linguistic forms. However, students from the unfocused group found the metalinguistic information difficult to apply in solving specific problems, suggesting that merely providing learners with metalinguistic explanations does not necessarily guarantee their deep understanding. These findings suggest that, when designing and conducting metalinguistic feedback, especially for EFL learners with limited proficiency, teachers must carefully consider the following issues: (1) whether the explanations contain difficult terminology; and (2) whether the examples are typical and clearly exemplify the targeted error category. To avoid of the problem of learners' overgeneralising the examples, a variety of examples can be provided to demonstrate various possible forms and corrections of the error. It is also desirable that the students be provided with opportunities to ask teachers questions to clarify their understanding of the metalinguistic feedback, as suggested by the literature (Chen et al., 2016;Ferris et al., 2013).

CONCLUSION
This study has found that coded and focused WCF with explicit metalinguistic support can significantly enhance writing accuracy in terms of the use of articles and singular/ plural nouns among L2 learners with low to intermediate language proficiency. A combination of circumstances, for example, the nature of the target grammatical structures and learners' understanding and ways of engaging with WCF, is also necessary for WCF to become a tool for learning.
Despite these insightful findings, some limitations must be acknowledged. First, the study was of short duration, and its sample was relatively small. Therefore, future research should investigate larger samples over longer periods of time. Second, although the three intact classes had a similar English proficiency level, the interviewees from the unfocused group had weaker proficiency than those from the focused group. Future qualitative studies should use participants with equivalent language proficiency to examine the extent to which language proficiency mediates the effectiveness of WCF.
Finally, this study has some implications for classroom pedagogy, addressed briefly here. Because not all errors deserve equal attention, teachers and students should consider how feedback can be used more effectively, particularly in areas where comprehensive feedback is considered obligatory. When teaching students with limited language proficiency, it is recommended that, rather than providing a wide range of error corrections, teachers provide focused feedback complemented with carefully designed metalinguistic support.

ACKNOWLEDGEMENT
The present article has received support from the College of Professional and Continuing Education, The Hong Kong Polytechnic University to cover partial expenses for the proofreading service.

DECLARATION OF COMPETING INTEREST
None declared.

ACCURACY (GRAMMAR, LEXIS, SPELLING)
Please check your accuracy carefully and correct the errors in the next draft.

Code
Meaning Example 1 VF Verb Form is wrong. VF: Take care of pets is an obligation of a pet owner.
Cor: Taking care of pets is an obligation of a pet owner.
VF: Billboard advertising presenting a good value for money.
Cor: Billboard advertising presents a good value for money.
VF: My priority is to focused on my career.
Cor: My priority is to focus on my career.
VF: I believe this solution should applied to Hong Kong by the relevant Hong Kong government department.
Cor: I believe this solution should apply to Hong Kong by the relevant Hong Kong government department.

Art
Article is incorrect, unnecessary, or missing.
(Article: a, an, the) Art: It was an humbling experience.
Cor: It was a humbling experience. Cor: Taking care of pets is an obligation of a pet owner.
VF: Billboard advertising presenting a good value for money.
Cor: Billboard advertising presents a good value for money.
VF: My priority is to focused on my career.
Cor: My priority is to focus on my career.
VF: I believe this solution should applied to Hong Kong by the relevant Hong Kong government department.
Cor: I believe this solution should apply to Hong Kong by the relevant Hong Kong government department.

Art
Article is incorrect, unnecessary, or missing.
(Article: a, an, the) Art: It was an humbling experience.
Cor: It was a humbling experience. Cor: I sprayed the ants in their hiding places.
WC: The dialectical interface between neo-Platonists and anti-disestablishment Catholics offers an algorithm for deontological thought. (Jargon or technical terms that are not suitable for readers) Cor: The dialogue between neo-Platonists and certain Catholic thinkers is a model for deontological thought.
WC: Another second way is to sound out staff. (Unnecessary words that lead to wrong phrases/structures) Cor: Another way is to sound out staff.

ACCURACY (GRAMMAR, LEXIS, SPELLING)
Please check your accuracy carefully and correct the errors in the next draft.

be
Be is incorrect, unnecessary, or missing Be.
This girl beautiful. Cor: This girl is beautiful.

Frag Frag
Fragment happens when a group of words lack a subject or a verb or fails to express a complete thought.
Frag: Since I missed last class. I did not know the homework.
Cor: Since I missed last class, I did not know the homework.

15
S-V Agr Subject-verb agreement s-v agr: There wasn't many students in class today.
Cor: There weren't many students in class today.

Others
Errors not listed above /