Introduction

In ^{1996, Truscott} argued that written corrective feedback (WCF) for errors can only contribute to explicit, rather than implicit, knowledge of a language and that this knowledge is irrelevant to actual language acquisition. Nonetheless, researchers have recently asserted that according to several theoretical approaches, we may expect writing (^{Manchón, 2011}; ^{Williams, 2012}), and moreover, WCF (^{Ferris, 2010}; ^{Van Beuningen, 2010}; ^{Bitchener, 2012}; ^{Polio, 2012}), to further second language acquisition. For example, according to skill acquisition theorists, systematic, repeated retrieval and use of explicit grammar rules develop automatized L2 knowledge, and WCF may hone this process. It has also been noted that WCF can be expected to be effective from the point of view of interaction second language acquisition (SLA) theorists, such as Merril ^{Swain (1995}), who have argued that output is a requisite for SLA as it enables learners to recognise gaps between their developing L2 systems and the target L2. Corrective feedback can draw students' attention to form and prompt noticing of these gaps. As WCF is given off-line, unlike oral feedback, which needs to be given during communication and therefore cannot be attended to so easily, it is even more likely to be detected (^{Ellis, 2003}; ^{Bitchener, 2012}; ^{Polio, 2012}). These theoretical bases help explain why recent studies of written feedback, which have focused on eliminating the methodological problems of earlier studies (identified, for example, in Ferris, 2003; ^{Bruton, 2009}), have found evidence for the positive effects of feedback on language acquisition (^{Ellis, Sheen, Murakami, and Takashima, 2008}; ^{Bitchener and Knoch, 2008}, ²⁰¹⁰; ^{Farrokhi and Sattapour, 2011}; ^{Hyland, 2011}; ^{Van Beuningen, De Jong, and Kuiken, 2012}; ^{Hanaoka and Izumi, 2012}; ^{Shintani, Ellis, and Suzuki, 2014}).

Despite the amassing evidence in favor of WCF for improving students' accuracy, there remains considerable doubt about which type of feedback may be the most effective (^{Storch and Wigglesworth, 2010}; ^{Hanaoka and Izumi, 2012}; ^{Bitchener, 2012}; ^{Ferris, Liu, Sinha, and Senna, 2013}). Since the majority of teachers do indeed give WCF (^{Lee, 2007}, ²⁰⁰⁸; ^{Furneaux, Paran and Fairfax, 2007}; ^{Guénette and Lyster, 2013}) and moreover, students wish to receive it (^{Leki, 1991}; ^{Ferris, 2003}), this question is highly relevant for teachers (^{Ferris, 2006}).

In general, written feedback may be direct, with a correct version supplied to the student, or indirect, in which case the presence of the error is indicated but not corrected. Within the indirect category, there are several possible subcategories depending on how explicitly the error type and location are indicated. The teacher may indicate the type of error committed through the use of a code (e.g. Gr = grammar); giving the location of the error, for example by underlining or highlighting it; simply noting the number or type of errors in a particular line of the text in the margin; or a combination of these (^{Robb, Ross, and Shortreed, 1986}). Table 1 gives some examples of the main types of indirect feedback:

Table 1 Different types of indirect feedback 

Some researchers have suggested that indirect feedback will better foster SLA since it requires learners to be more active in their response to it, i.e. as they apply their existing knowledge to solve the problem of correcting the indicated error (^{Ferris et al., 2013}.). One study that supports this was carried out by ^{Storch and Wigglesworth (2010)} . Students composed a text in pairs, received direct or indirect feedback on it, and discussed that feedback together. Those pairs that received indirect feedback were more likely to discuss the feedback points (producing more "language-related episodes") than those that received direct feedback. However, it has also been speculated that direct feedback may be more beneficial for SLA, provided it is delivered in a manner which is in line with the writer's intended meaning. The reason posited for this greater effectiveness of direct feedback is that it provides unambiguous, immediate information about the correct version, thereby enabling learners to notice the gap more efficiently between their current performance and the target feature (^{Bitchener and Knoch, 2010}, ^{Ferris et al., 2013}). Moreover, it avoids possible difficulties in deciphering and utilizing codes to modify their drafts (^{Ferris, 2003}).

Empirical studies into the effectiveness of the different types of feedback have failed to produce any clear conclusions to date. Early studies did not show any significant difference in improvement in writing accuracy (e.g. ^{Lalande, 1982}; ^{Semke, 1984}; ^{Robb et al., 1986}; ^{Chandler, 2003}), but were plagued by methodological issues such as variation in the genre of the elicitation tasks; individual student differences (age, motivation, etc.); the amount of writing done; the length of treatment; other classroom activities (e.g. amount of grammar instruction); the number and type of grammatical errors focused on; and post-test measures (type and delay) (^{Ferris, 2003}; ^{Bruton, 2009}). The few more recent studies involving tighter methodological controls have obtained sounder, significant results at the delayed post-test level, but the results have been inconsistent: ^{Van Beuningen et al. (2008}) and ^{Bitchener and Knoch (2010}) found direct feedback to lead to greater improvements than indirect feedback, whereas ^{Van Beuningen et al. (2012)} found that accurate use of grammatical forms improved with direct feedback, but accurate use of non-grammatical forms improved with indirect feedback; the authors were not sure why this difference should exist.

Another area that has been of interest to researchers looking at different types of feedback regards students' preferences, but studies of preferences have also drawn mixed responses. ^{Ferris (2003}), for example, found students preferred indirect, coded feedback over direct feedback, yet students in ^{Lee's (2008}) study favored direct feedback, and those in other studies have shown relative ambivalence (^{Saito, 1994}; ^{Ferris and Roberts, 2001}).

The ambiguity which marks the debate over the effectiveness of feedback types probably reflects the fact that a number of different factors affect student responses to feedback. One important factor may be proficiency. ^{Bitchener (2012}) has posited that since "learners at a lower level of proficiency ... may not have such an extensive or deeply processed linguistic knowledge base to draw upon" (p. 355), direct feedback is likely to be of more benefit to them. Learners' language analytic ability and previous level of grammar instruction are perhaps even more likely to be of relevance than overall proficiency: ^{Sheen (2007}) has already demonstrated that learners with better analytic ability benefit more from direct feedback, and it may be expected that they would benefit even more from indirect feedback, which requires some use of that ability. Similarly, learners with strong metacognitive knowledge of grammar are more likely to be able to take advantage of indirect, coded feedback. Learners with little or no such knowledge may be expected to benefit more from direct feedback, or possibly indirect, uncoded feedback (see types 1 and 2 in Table 1) (^{Bitchener, 2012}; ^{Ferris et al., 2013}). There is relatively little evidence, though, to support these hypotheses. Some evidence of how students with weak metacognitive grammatical knowledge had difficulties with coded feedback can be found in the Ferris et al. (2013) study of ten university-level students in the US. The researchers tried using indirect, coded feedback with these Generation 1.5 learners in the US (children of first generation immigrants, who were born in the US or arrived at a young age, and therefore were highly proficient) who had learnt most of their L2 through immersion, but had had some grammar instruction at school. These learners preferred to rely on intuitive linguistic knowledge in order to deal with the WCF. When they did try to use the grammar instruction they had received at school, that instruction was not often well remembered and hence not particularly helpful.

As well as students' proficiency and linguistic knowledge, the teaching context may well be relevant in determining students' preferences for different types of feedback. For example, ^{Lee (2008}) judged that Hong Kong high school students' preference for direct feedback had been shaped by the teacher-dominated nature of lessons with minimal space for conferencing or discussion of errors. Student perceptions of their learning needs (areas of difficulty), and whether or not the teacher recognises and meets those needs in their feedback (^{Hyland, 2003}; ^{Storch and Wigglesworth, 2010}; ^{Ferris et al. 2013}) can be expected to strongly shape students' responses to feedback.

In summary, the focus in publications on WCF has moved recently to highlight theoretical grounds for expecting WCF to be effective, and research has begun to show consistent patterns to that effect. There is, however, considerably less clarity regarding how students use and perceive different types of feedback, an area of important interest to teachers. I decided to conduct the present study as I believed that my students were not finding the direct WCF I was providing useful, and decided to explore using indirect feedback whereby I indicated the type of error made in the margin (type 3 in Table 1 above). The study was therefore carried out within an action research framework. My main research question was: How would my students respond to the direct and indirect feedback? As well as contributing to the indirect vs. direct feedback debate, this study responds to calls for longitudinal research on WCF that is carried out in genuine teaching-learning contexts, rather than contrived experimental situations (^{Hyland, 2003}; ^{Ellis, 2010}; ^{Storch, 2010}, ^{Ferris 2010}; ^{Van Beuningen et al., 2012}) and, relatively unusually, focuses on writers in a non-English speaking country (^{Ferris et al., 2013}; ^{Lee, 2014}).

Method

Participants and context

The learners in this study were undergraduates on a variety of programs taking a year-long optional English course at a university in the North of Chile. The class comprised ten students, six of whom completed the course and participated in this study. All the participants were females in their twenties, and their first language was Spanish. They had learnt English through a mixture of classroom-based learning and immersion experiences. Three of the students had passed a pre-intermediate course at the same institution the previous year, which followed a similar structure and assessment pattern to their current course. These students had received direct WCF on their written work from the same teacher during the pre-intermediate course. Apart from this experience, all the participants claimed they had not received WCF on any previous written work. The participants' English level was generally close to "upper intermediate," level 3 of the Association of Language Testers of Europe, as attested by their performance in the University of Cambridge First Certificate of English (FCE) examination at the end of the course. The participants were highly motivated, as evidenced by their persistence in this optional course and eagerness to pass their pending international examinations. A summary of their profiles is given in Table 2 below:

Table 2 Profiles of the participants 

The English course lasted two semesters and aimed to improve students' proficiency in English and prepare them for the FCE examination. Students attended five hours of classes per week. Two hours of these were taught in a single "theory" class that focused on grammar and metalinguistic knowledge, which were taught deductively and practised using a course book. The remaining three hours were divided into two one-and-a-half hour classes that focused on oral communication. Over the year, students submitted eleven written homework assignments, each approximately 200-250 words in length, for their theory class. Assignments varied in genre and included, for example, stories, reports, and letters. They were taken from the course textbook and graded according to a rubric.

All of the students consented to participate in this study. They were informed that their participation was voluntary and would not affect their grades, that their names would be changed in any written report of the study, and that they were free to withdraw at any stage.

Feedback provision

This research was based on an action research project, i.e. it involved a planned intervention in response to a perceived problem with a systematic collection of data to evaluate the intervention (^{Burns, 2005}). As explained above, prior to the study, I provided students with direct feedback on their homework. I had done this for their first seven written homework assignments (for an example, see ^Appendix). Before returning the assignments to the students, I noted common errors on the board. Students were asked to correct the errors and they were subsequently discussed. The assignment was then returned, and students were given time to check the feedback and ask questions. The students were not required to submit another draft of this work. Over time, I became concerned that despite their motivation, my students were not benefitting from this feedback as it required little reflection from them; they simply looked at it and put it away, usually without asking questions about the corrections.

As the literature suggested that indirect, coded feedback would call for deeper cognitive processing (^{Bitchener, 2012}), I decided to try using it with my students for their remaining four assignments. Since my students had some metacognitive knowledge of grammar from their theory classes, I opted to use coded feedback for "treatable" errors (^{Ferris et al., 2013}). Following ^Ferris' suggestion ⁽²⁰⁰⁶⁾, I would continue to use direct feedback for "untreatable" errors, i.e. errors that I felt they would be unable to correct without support due to their complexity; these usually pertained to word choice or more complex sentence structure issues. Furthermore, as I wanted to encourage my learners to reflect on their work as deeply as possible, I decided not to not to indicate the exact location of errors, but rather to note the presence of errors in the margin (see Table 1, Feedback Type 3). An example of how this feedback looked in practice would be:

Gr He usually wears jeans when he went to his job work.

I would use 'Gr' in the margin to indicate the wrong tense used in "wore" as students would be expected to know this from the course, but would give direct feedback to the use of "job" as the difference between the two words was not one we had studied (see ^Appendix for an example of feedback given to a more extended text).

Three weeks into the second semester, I told my students that I planned to change the type of feedback I provided and that I was interested in knowing their responses to the two different types of feedback. I explained how I would now provide feedback and that class time would be given for checking this feedback. Students would have the opportunity to amend their errors and submit a second draft the following week. If errors remained on the second draft, direct feedback would be provided for those. Students' final grades for the assignment would be an average of the grades for the first and second drafts. If they did not submit a second draft, the grade from the first draft would stand. Table 3 shows which homework assignments were submitted by the different students:

Table 3 Written homework assignments submitted 

The coded feedback was adapted from ^{Ferris (2006}) and included codes for 12 error categories, such as T -Tense, C -Conjugation, P -Punctuation, etc. The codes were trialled on photocopies of two previous assignments to check whether they needed modifying. Before assignments with coded feedback were returned for the first time, a code sheet, which gave an explanation and example for each symbol, was discussed with the students.

The feedback given was analysed in order to verify that the two types of feedback were provided as intended, which was important since intentions to use a particular type of feedback do not always translate into practice (^{Ferris, 2003}), and that both types of feedback were delivered relatively consistently. Consistent feedback would mean that I omitted to give feedback (by mistake) to an error, or gave misleading or inaccurate feedback, to a similar proportion of errors regardless of the type of feedback being given.

In order to analyse the feedback provision, counts of errors and direct and indirect feedback points were taken from the four assignments that received coded feedback, and the final four assignments to which the teacher had given only direct feedback, in order to have a comparable sample (i.e., assignments 4-7 and 8-11). The counting of errors and feedback points was carried out as follows. First, the number of words written for each assignment was counted. Second, for direct feedback essays, the number of errors from each error category was counted. Of these, any errors that had not received direct feedback, or that had been corrected inaccurately (i.e. the correction was simply wrong, or appeared to change the meaning of the original text), were identified and counted. Third, for the first draft of the indirect feedback essays, treatable errors (i.e. those that students could be expected to be able to self-correct, based on the material already covered in class and knowledge of the student's level) were counted. Instances where treatable errors were not given feedback, were given incorrect feedback (e.g. through the use of an incorrect code), or were given direct feedback, were also counted.

The words and errors were counted independently by two raters in order to verify counts. The first rater was the teacher/researcher; the second was a British teaching assistant. Once completed, the raters compared their ratings to reach 100% agreement (as in ^{Ferris, 2003}). Normalised scores for the feedback points were calculated using Biber, Conrad and Reppen's (1998) procedure (cited in ^{Ferris and Roberts, 2001})¹.

The error counts for the two types of essay indicated that a very low percentage of errors were given incorrect feedback on the indirect and direct feedback essays: 1% and 2%, respectively. The proportion of errors that were not given feedback on the direct and indirect feedback essays was the same (14%). The data are comparable with findings from a study by ^Ferris on the basis of which she concluded that the "feedback was by and large both comprehensive and accurate" (²⁰⁰⁶, p. 92). It seems reasonable to conclude, then, that any differences perceived between the two types were unlikely to be due to differences in quality of provision.

Findings

When the indirect feedback was first introduced, students, perhaps unsurprisingly, took time to get used to it. As mentioned above, on the first occasion, I discussed the code sheet with the students before returning their work and then gave them some time to look at the feedback and ask any questions. I noted in my journal that on this first occasion, most students spent several minutes checking the codes and asking questions, but as the weeks went by, they used less time for this. The students confirmed in their interviews that although they had initially found the indirect feedback confusing, they soon became used to it. In Andrea's words:

At the beginning I said, 'oooh... how complicated,' ha ha ha, because I said, 'so many forms of seeing errors, the prepositions, the punctuation, I don't know, the tenses.' At the beginning it was difficult but then I saw it was better.

This perception that they were able to utilise the indirect feedback was echoed in the quantitative analysis of whether students incorporated the indirect feedback in their second drafts: on average, 72% of the indirect feedback was incorporated according to the code given, and in 5% of cases, the text was modified in a different but appropriate manner. Overall, 77% of the coded WCF was incorporated effectively. Once again, these findings are similar to those from ^{Ferris's (2003}) study. The incorporation scores can be seen in Table 4 below:

Table 4 The number of feedback points incorporated by students in their second drafts for the coded feedback essays according to type of feedback given 

As can be seen from Table 4, Beatriz's score was particularly low at 36%, which can be explained by the fact that Beatriz submitted only one second draft, and the first draft of that essay had been problematic in terms of its overall structure. She modified it extensively in the second draft and therefore did not appear to have incorporated much feedback. Beatriz expressed commitment to using the codes in the interview, and it seems fair to speculate that had she produced more second drafts, her incorporation score would have been higher.

The other student who also utilised the indirect feedback to a lesser extent was Maria (46%). Maria wrote in her questionnaire that she would have preferred to have only direct feedback and that she felt she "wasted time" trying to correct the coded feedback. Unfortunately, it was not possible to interview Maria to ask her opinion in more depth. However, her learning background may explain her dislike: Maria learned most of her English on an exchange programme in the US (see Table 2), and was one of the most proficient speakers in the group. I knew from my interactions with her in class that she enjoyed activities that were more communicative in nature and did not enjoy the types of activity that required attention to detail or practising grammar. This was reflected in her result for the length, demanding First Certificate exam, in which she obtained a D, despite being having, in my opinion, the necessary proficiency to obtain a much higher grade. In many respects, Maria's response to the indirect feedback echoes that of the Generation 1.5 learners in ^{Ferris et al.'s (2013}) study: students who have learned their English through an immersion-type setting and who dislike and/or have little metacognitive knowledge of grammar are unlikely to respond positively to indirect feedback.

Another area of difficulty that was touched on by one student regarding the indirect feedback was frustration at feeling uncertain about whether the correction of an error was indeed correct or not, and not having an immediate answer to any hypotheses tested out; ^{Bitchener (2012}) notes that these are supposed advantages of direct feedback. Carmen mentioned her frustration during the interview: "Sometimes I'm very unsure so, for example, I don't know, uh... it said that the verb form used was incorrect, so there were various verb forms, so which of all of them is the right one?" Although, as mentioned, I gave the students direct feedback on any errors that remained in the second drafts, it may have been have been helpful to encourage these learners, and Carmen in particular, to check whether their corrections had been successful (for example by giving more class time for this).

Despite these negative responses to the indirect feedback, most students recognised that it had a number of advantages. One of the most prominent benefits that was perceived was that it required a more active response from learners. This perception was not based simply on the fact that they now had to incorporate the corrections in a new draft (which they had not done with the direct feedback). As Carmen said: "You realise what is wrong, uh, more than simply being told what is wrong, like you have to make more effort to find what is wrong."

Furthermore, the students made a direct connection between making more effort and learning more from the coded feedback. For example, Beatriz reported: "[The indirect feedback] was better because it made me put into practice and think a little more in what, in what the error had been the first time I did it. [...] The learning stuck, it stuck because it was deeper." Similarly, Diana criticised the direct feedback as being too passive and hence ineffective: "I think that [direct feedback] made the learning less meaningful, since it was limited to correcting the error in the way shown." These students clearly believed that the indirect feedback helped as it drew deeper cognitive processing (^{Van Beuningen, 2010}; ^{Bitchener, 2012}). From the data collected, it is not possible to know whether this may have led to better writing or self-monitoring skills, or indeed actually have contributed to acquisition (implicit knowledge). However, as ^{Van Beuningen (2010)} and ^{Polio (2012}) have argued, according to several theories of learning -including usage-based, sociocultural, and interactionist approaches- paying attention to feedback is likely to be beneficial for language acquisition, as well as improved accuracy. Five of these students certainly perceived that the indirect feedback caused them to pay greater attention to their language use.

Another advantage of indirect feedback perceived by the students that emerged from the data was that it helped to reinforce their grammatical knowledge. Four of these students had learnt their English prior to the course at school in a typical EFL context and, as mentioned, in the English course they were taking, two out of five hours per week were focused on grammar. Moreover, this was a context where the emphasis was on general English proficiency, not one specifically directed a learning to write. ^{Ferris et al. (2013}, p. 309) posited that students in EFL contexts who have received considerable formal grammar instruction "may benefit from WCF that includes specific terms or rule reminders, as the codes, corrections, or explanations may elicit their prior knowledge." These students had not only received a considerable amount of grammar instruction; they also believed that it was highly beneficial. In their questionnaires, five of the students thought that grammatical knowledge was as important as, or more important than, organization and content to improve their writing. In response to an interview question about what had been the most important factor for improving their writing, Bianca answered: "The grammar, by far the grammar, because if before I knew how to write something I didn't know why, and now I understand why." Beatriz also thought that grammatical knowledge was helpful in improving her writing: "Once I understood the rules it was much easier and so like in maths you apply them pum-pum and it comes out right."

Some of the students did explicitly mention, as ^{Ferris et al. (2013}) predicted, that the indirect feedback reinforced their grammatical knowledge. Carmen, for example, stated that she liked the coded feedback because it helped her: "To know more specifically which errors I typically commit, so as to improve that aspect of the grammar."

Another advantage of the indirect feedback that I had not anticipated was that it seemed to foment autonomous learning behaviours. Andrea, for example, sought me out in my office on more than one occasion (as noted in my journal) to discuss difficulties she was having. She was also motivated to look up information: "There were times when some of [the errors] were very difficult to understand, I admit that, so that then, well, sometimes I started to look in texts, sometimes in the same book we used." Similarly, Diana commented that it led her to return to material studied in class: "In the process of discerning what my error was, I had to review the topics already taught."

The indirect feedback particularly suited those of my students who were generally highly motivated learners. Andrea, Carmen, and Bianca went to great lengths to incorporate as much indirect feedback as possible, as is apparent from their incorporation of more than 85% of it in their second drafts. When I asked them in their interviews why they had worked so hard on this, their determination and satisfaction were clear:

Carmen: Hmm... because I am a PERFECTIONIST, ha ha ha. Because I like the details of things to be right, I really notice the details, so uh, so that's why I gave it so much attention.

Bianca: In general in student life I have always done it like that. When I have a wrong answer I go and I look in my notes for the part that I got wrong, or when I come out of a test and I wasn't sure of what I was writing, I always look for the answer, and that information you definitely never forget it.

To summarise these findings, I will consider the students' overall preferences for feedback, as expressed in their questionnaires. All of the students indicated that they would prefer to receive some form of WCF rather than none. The main reason given was that they would have been unable to identify errors without this support (similar to students in ^{Lee, 2007}) and improve in the future. As Carmen explained:

Because if you're not criticised and [the teachers] find everything good, then you never know if you learn or not. I mean I like it when they tell me, "No, it's wrong, or correct it, it could be better." I like that because I believe that way you can improve.

When asked what type of feedback they would wish to continue with in the future, Maria and Carmen responded that they would prefer direct feedback; Maria perhaps due to her learning preferences, and Carmen due to the difficulties in knowing which answer was correct. The remaining four students stated that they would prefer indirect, coded feedback. Of these last four, two noted that they would have preferred to have their errors underlined as well as coded. With hindsight, I believe that underlining the errors may help to make the revision and correction process more manageable for students.

Limitations, conclusions and implications

Before considering the conclusion that can be drawn from this study, it is important to recognise the methodological issues. As with any case study, the sample was small, and of those students that did participate, not all completed all of the essays and not all were available for interview. Moreover, although there was some direct feedback (for treatable errors) on the essays that were rewritten, a fair comparison would have been possible if students had also had the chance to rewrite the direct feedback only essays. The fact that I, the teacher, carried out the interviews and questionnaires may also have affected how students expressed themselves.

Despite its methodological drawbacks, this paper presents data that respond to calls for ecologically valid evidence from a long-term study of students' responses to different feedback types in a genuine EFL teaching context (e.g. ^{Ferris et al., 2013}; ^{Lee, 2014}). Indirect feedback is sometimes questioned on the basis of whether learners may be able to use it (^{Bitchener, 2012}), but the data collected showed that these learners were generally able to use the indirect feedback to modify their writing. In this university context, students who were motivated and valued metacognitive knowledge of grammar found indirect, coded feedback to be more helpful as amending it became a more active process, and they believed this helped their learning. Some specifically found the indirect feedback helped reinforce their grammatical knowledge, which they believed to be helpful for their writing, as proposed by ^{Ferris et al. (2013)}. For a few students, the indirect feedback was as a springboard for positive autonomous learning behaviours, such as checking reference books. Two of the students, however, considered the direct feedback to be more useful. In one case, this was because the student felt it was difficult to know how to respond to the feedback correctly; the other student, who had learnt her English mostly through an immersion experience, found the indirect feedback to be unhelpful (see also ^{Ferris et al., 2013}, for example).

The data collected therefore strongly suggest that the grammar-oriented EFL teaching context and the students' previous learning experiences and levels of motivation affected the students' responses to the different types of feedback. It reinforces the notion that a "one-size-fits-all" approach for feedback provision is unlikely to be suitable; rather, there is a need for teachers to be aware of the learning context and individual differences when making choices in this area (^{Hyland, 2003}; ^{Storch and Wigglesworth, 2010}). Where possible, it would seem recommendable that teachers offer students different types of WCF and allow them choice regarding which type to use, especially given that learner beliefs about how learning happens are one of the key factors in making learning happen (^{Cotterall, 1995}; ^{Benson and Lor, 1999}).

A need remains for more research to clarify which type of WCF, including different types of indirect feedback, may be most effective, with which types of students, and why. Several authors (e.g. ^{Bitchener, 2012}; ^{Polio, 2012}; ^{Ferris et al., 2013}) have suggested how factors such as learning context, proficiency levels, and metacognitive knowledge of grammar may impact responses to different types of feedback; this study is one of few that have begun to explore these factors in practice, and it would be useful to continue to develop this knowledge base. There is a need, too, for more research in authentic EFL contexts, and to continue to consider learners from the wide variety of contexts where EFL writing takes place.

Thanks

The author would like to thank Fiona Hyland for her helpful comments on earlier versions of this article.