Using Models in Writing Instruction

Models are commonly employed in both first (L1) and second language (L2) writing classrooms; however, questions remain concerning the role of this inductive technique in students’ genre learning. The study reported here examines the effect of two instructional techniques (models and models combined with explicit instruction) on the ability of three groups of writers (native speakers of English, higher proficiency nonnative speakers, and lower proficiency nonnative speakers) to produce a specially constructed essay type. Using a controlled/posttest design with stimulated recall data, the study found that those students receiving models were outperformed on Essay 1, Essay 2 (1 week later), and a Quiz (1 month later) by those students who received models in combination with explicit instruction. Implications for L2 writing instruction are discussed.


Introduction
For novice writers, learning a new genre can be a daunting experience: not only do they have to learn the conventions associated with that genre (e.g., strategies for engaging and interacting with the reader, signaling confidence in claims, and indicating authorial presence), but they also need to understand the sociorhetorical reasons for the conventions' use and ultimately gain the ability to appropriate those conventions for individual creative and rhetorical purposes (e.g., Bawarshi & Reiff, 2010;Johns et al., 2006). To assist novice writers in this task, teachers of both first (L1) and second language (L2) students have long made use of model texts (e.g., Charney & Carlson, 1995;Hillocks, 1986;Macbeth, 2010;Stolarek, 1994).
The use of these models, however, has evolved considerably over time. In the 1960s and 1970s, models were valued for the opportunities they provided students to imitate the "correct" forms and for the tight control they provided over the learning process (e.g., Eschholz, 1980;McCampbell, 1966;Paulston, 1972). For example, students would be presented with a model text and instruction on its organizational or grammatical features; they would then be asked to complete a series of controlled activities (e.g., changing the model from the present tense to the past or rewriting it so as to address another topic) before engaging in more independent writing activities. Criticisms of this approach began to mount in the 1980s, however, as theorists argued that models were often overly prescriptive, denying students their individual voices and creativity. In addition, the use of models was said to undermine the writing process by privileging form before the development of ideas and by failing to provide information to students on the writing processes that gave rise to the models (e.g., Collins & Gentner, 1980;Murray, 1980;Taylor, 1981;Watson, 1982;Werner, 1989;Zamel, 1983).
In current approaches to writing instruction, models are valued not for facilitating passive imitation, but rather for (1) raising the visibility (and accessibility) of target rhetorical conventions (e.g., Hyland, 2003Hyland, , 2004; (2) helping students create a mental model of the genre (e.g., Crinon & Legros, 2002); and (3) easing some of the apprehension associated with writing a new genre (e.g., Macbeth, 2010). For example, in the genre-based approach to writing instruction, multiple models of a particular genre are analyzed with respect to their organizational, lexicogrammatical, and rhetorical features. Students are sensitized to the genre's social context and to a range of methods for addressing the needs and expectations of particular disciplinary audiences. This analysis serves as a foundation for joint construction of the genre 426295S GOXXX10.1177/21 58244011426295AbbuhlSAGE Open 1 (e.g., teachers and students working together to produce examples of the genre) and later individual construction (e.g., Feez, 1998;Hyland, 2003Hyland, , 2004. For adherents of such an approach (e.g., English for Specific Purposes or ESP scholars), models are useful for supplementing explicit instruction. Models alone, although perhaps sufficient for "middle class L1 students who [are] immersed in the values of the cultural mainstream [and] share the teacher's familiarity with key genres," are considered insufficient for L2 writers (Hyland, 2003, p. 19). As Hyland (2003) notes, L2 writers typically are not aware of the "patterns and possibilities of variation" (p. 19) across genres; as such, they are unlikely to benefit from inductive approaches (such as models alone) that do not include consciousness-raising activities to draw learners' attention to the target rhetorical features. However, this approach to models and instruction has not escaped criticism. For example, it has been claimed that instruction (and its use of models) is potentially reductionist, ignoring the complexity/dynamicity of genres and the ways in which organizational and rhetorical features can be manipulated for individual purposes (e.g., Bawarshi & Reiff, 2010;Cope & Kalantzis, 1993;Johns, 1995;Prior, 1995). Models and their accompanying instruction have also been criticized for perpetuating genres of power that exclude L2 students (e.g., Benesch, 2001). At best, it is said that modeling and instruction may lead to genre competence (the uncritical ability to recognize and reproduce key features of the genre) but not to genre performance (a more sophisticated ability that allows for a critical appraisal of the target features and the ideologies they embody as well as the ability to appropriate features for individual rhetorical purposes; e.g., Bawarshi & Reiff, 2010). For Rhetorical Genre Study (RGS, or New Rhetoric) scholars, immersion-based approaches (e.g., disciplinary apprenticeships) are the driving force behind genre performance (e.g., Bazerman, 1988;Berkenkotter & Huckin, 1995). This view is not incompatible with explicit instruction (see, for example, Bawarshi, 2003), but some RGS theorists have claimed that explicit instruction is not necessary for genre learning, and additionally, that models need not cooccur with explicit instruction (e.g., Freedman, 1993, p. 248, argues that "simply by reading the models, students may be inferring the relevant rules by themselves, at a level below the conscious").
Despite these ongoing discussions, however, there is very little empirical research related to the effectiveness of models and whether they need to be supplemented with explicit instruction. Although there are a few studies addressing this question in L1 writing classrooms, research on models in L2 writing classrooms is scarce. If the field is to make informed decisions about the use of models for L2 writers and address the equally important (and under-researched) question of whether explicit instruction is necessary or useful for L2 genre learning (e.g., Bawarshi & Reiff, 2010;Paltridge, 2007;Polio & Williams, 2009), then we cannot rely solely on evidence from L1 students but rather build a body of literature that is based on the populations we teach.

Literature Review
There are only a handful of studies in the L1 literature that have sought to examine the effectiveness of models. In one of the earliest, Bereiter and Scardamalia (1984) examined the effectiveness of models in the genre learning of suspense stories, restaurant reviews, and an artificial genre they named "concrete fiction." They found that their participants, L1 students ranging from third graders to graduate students, benefited from the models, but only concerning "discrete elements of language and content rather than toward more global aspects of form or rhetorical strategy" (p. 177). Generally positive results for models were also reported in Charney and Carlson (1995). In this study, the researchers examined the ability of undergraduate students to write the methods section of a short experiment. Comparing the control group to the treatment groups (who read either three good models or three models of varying quality), the researchers found that those students who received the models had better organization and included more necessary details in their texts (although they also included more unnecessary details) than did the students who did not receive models. More recently, Crinon and Legros (2002) found that the use of models helped their 8-to 10-year-old participants with respect to the amount of details they included in their stories.
However, less positive findings have also been reported. For example, Hillocks' (1986) meta-analysis of 60 experimental studies conducted between 1963 and 1982 found that models had a weaker effect on the quality of students' writing than other forms of instruction. Smagorinsky's (1992) comparison of three treatments (models alone, models with instruction on the composing process, and models with instruction on the writing of extended definitions) led to the conclusion that "reading models alone is insufficient to improve writing" (p. 173). Most recently, Stolarek (1994) used Bereiter and Scardamalia's (1984) artificial genre to investigate the effects of models on university freshmen (n = 143) and composition instructors (n = 21). Participants were placed into one of five groups: (a) description only, (b) model only, (c) description and model, (d) model and explication, and (e) description, model, and explication. Looking at a range of features (including adherence to the two-paragraph format, use of third person, and inclusion of detail), Stolarek found that the students did poorly in both the description only and model only conditions, but when models were combined with description and/or explication, the students performed similarly to the composition instructors.
Studies on the use of models with L2 learners are considerably scarcer. Although a number of case studies have mentioned that L2 writers make use of models (e.g., Angelova & Riazantseva, 1999;Casanave, 1998;Leki, 1995;Tardy, 2009) and studies on reformulation (a type of feedback) have noted that L2 writers can use models to address their lexicogrammatical errors (e.g., Qi & Lapkin, 2001), only one study to my knowledge (Henry & Roseberry, 1998) has compared the use of models with other instructional techniques.
In Henry and Roseberry (1998), the researchers compared two forms of instruction involving model tourist information texts. The genre group read six examples of the target genre and discussed the typical moves (organization) of the genre, as well as some of the more common lexicogrammatical features. These students then rewrote a flawed example of the genre and created their own tourist information text. The nongenre group read the same model texts, rewrote the flawed genre, and produced their own tourist information text. However, whereas the genre group's attention was drawn to the moves of the target genres, the nongenre group completed more traditional grammar exercises. Examining the two groups' gain scores (differences between pretest and posttest), the researchers found that the genre group made significant improvements in the cohesion and persuasiveness of their texts, with improvements in organization approaching significance. The nongenre group's gain scores did not reach statistical significance. Although the authors did not discuss the results in terms of models, this finding does suggest that models may need to be supplemented with explicit instruction for L2 writers.
However, more recent work has suggested that models, while not sufficient for genre learning, do provide less proficient L2 writers with the support they need to generate such aspects of a text as thesis statements, topic sentences, and support sentences (Macbeth, 2010). In her qualitative study, Macbeth found that "for the most insecure writers, the skeleton was the most visible set of instruction on how to write an essay, and ultimately it offered a surface to which further instruction could cling" (p. 45)-a finding that led her to conclude that models may help students navigate the difficult terrain of an unfamiliar genre. However, Macbeth also noted that students eventually have to recognize the reduced nature of models so they are not locked into a simplified and artificial style of writing.
Given the scarcity of research (and the ubiquity of models in L2 writing classrooms), it would seem imperative to investigate the effect of models on L2 writing and whether models need to be supplemented with additional instruction. In the present study, the following research questions are addressed: 1. Do (a) models and (b) models combined with explicit instruction have a differential effect on students' ability to produce specially constructed essay types? 2. If so, does language proficiency (native, nonnative higher proficiency, and nonnative lower proficiency) mediate that effect?

Method Participants
There were 135 participants in this study, all currently enrolled students at an American university. Due to participant attrition, the final sample size was 118. Of the final sample, 41 participants were native speakers (NS) of English (operationalized as individuals who began learning English prior to the age of five and who had completed their K-12 schooling in English; note that it was not necessary for an individual to be a monolingual to be classified as a native speaker). The remaining participants were nonnative speakers of English (NNS, operationalized as individuals who began learning English after the age of five and completed their K-12 schooling in a language other than English). These 77 NNS were divided into two proficiency groups based on two criteria: (a) whether they had passed the Writing Proficiency Examination (WPE, a university-level 75-min timed essay test that students need to pass with a score of 11 or higher before graduating) or equivalent (such as the analytical writing section of the Graduate Record Examination) and (b) whether they were currently enrolled in remedial writing courses. The "higher proficiency" NNS (n = 40) had received a passing score on the WPE or equivalent and were not currently enrolled in remedial writing courses; the "lower proficiency" NNS (n = 37) had received a nonpassing score on the WPE or equivalent and/or were currently enrolled in remedial writing courses. 1 Having been recruited by flyers posted around the university and entered into a cash drawing for completing all portions of the experiment, participants in each group were diverse with respect to their majors (e.g., Asian American studies, biology, business, economics, engineering, history, linguistics) and first languages (23 different languages, including Arabic, Chinese, Farsi, Japanese, Korean, Russian, Sinhala, Turkmen, and Vietnamese). Questionnaire data indicated that the students' perceived writing abilities were also varied.

Materials
To investigate the effect of instruction and models, one of course must choose a target genre. Previous studies on explicit instruction and genre learning have employed, among others, tourist information texts (Henry & Roseberry, 1998), term papers (Mustafa, 1995), and job application letters (Henry, 2007). However, one possible confounding variable is that of prior exposure: if participants have previous experience either reading or writing the target genre, then it becomes difficult to isolate the effect of instruction (Tardy, 2006). When the subject population is relatively homogeneous, it is possible to choose a genre that no participant has had prior exposure to; however, when the participant pool is as heterogeneous as it is in the present study, then choosing a target genre becomes more problematic. To control for the effect of prior exposure, previous researchers have created artificial "genres" or essay types (Bereiter & Scardamalia, 1984;Stolarek, 1994), akin to the miniature artificial languages that researchers have developed to investigate the acquisition of L2 syntax.
However, there is a problem with relying on a single artificial essay type, and that is shared accidental similarity with participants' preferred or previously learnt styles of writing. For example, if a single essay type is employed and one group (say, the native speakers) outperforms all others, attributing the native speakers' performance solely to the instruction or models becomes problematic as there is a possibility that the target essay type resonated with the native speakers' preferred/previously learnt styles of writing. Thus, in the present study, two artificial essay types were created, the "Type A" and "Type B" "Translantian three-paragraph persuasive essay." (Both had been piloted in a previous study.) The name of a fictional country, Translantia, was chosen to help participants understand that the essay types were artificial. The essay types were kept short (approximately 140 words), as previous researchers have suggested that using long, complex genres in laboratory settings is problematic (Smagorinsky, 1992).
The Type A essay type made use of personal pronouns to signal authorial presence, boosters (words and phrases that convey the writer's certainty about a proposition, for example, it is certain, obviously), linking adverbials associated with speech (e.g., so, besides), rhetorical questions, and a three-paragraph style of organization (description, personal anecdote, and opinion). The Type B essay type made use of third-person pronouns and phrases (e.g., the author) to signal authorial presence, hedges (words and phrases that signal the writer's lack of full commitment to a particular proposition, e.g., it seems, possibly), linking adverbials associated with writing (e.g., therefore, in addition), directives, and a threeparagraph style of organization (personal anecdote, description, and opinion). Hyland's (2005) criteria were used for identifying hedges and boosters, and Biber, Johansson, Leech, Conrad, and Finegan's (1999) guidelines were followed for distinguishing linking adverbials associated with speech and those with writing. Two examples of each essay type were created, and each example was approximately 140 words long. Each example contained five signals of authorial presence, five markers of epistemic modality, five connective devices, and three reader engagement signals (three rhetorical questions in the Type A essay type and three directives in the Type B essay type). The differences between the two essay types are summarized in Table 1 and examples can be found in the appendix.
It must be noted at this point that generalizability becomes a concern when using artificial essay types. In using an artificial essay type, both the audience (in the present study, the Translantians, fictitious residents of Translantia) and the communicative purpose (here, persuading native Translantians) become a fiction. As authentic genres are multidimensional, dynamic, and socially situated (Bawarshi & Reiff, 2010;Flowerdew, 2011), questions must be raised concerning the extent to which the results from a study on artificial essay types are generalizable to learners encountering authentic genres in rich social settings. This is a legitimate concern and one that will be addressed in more detail in the conclusion.

Procedure
There were 12 groups in the present study. First, students were randomly assigned to the "instruction" or "no instruction" group. Students in the instruction group received models of the target essay type and explicit instruction on the models' rhetorical features. The students in the noninstruction group received only the models. Students were then assigned to the Type A or Type B essay type groups, yielding the distribution of students given in Table 2.
All students first completed a background questionnaire on their writing/reading habits and their perceived English writing abilities. All students were then informed that they would be learning how to write a new style of persuasive essay from the fictitious country of Translantia. Students were told that they would read four short examples of Translantian persuasive writing, two "effective" and two "ineffective," and that they should pay careful attention to the style of writing employed in the "effective" essays as they ultimately would be writing an essay for a native Translantian. Both "effective" and "ineffective" essays were used as researchers have suggested that the juxtaposition of the two can enhance the salience of rhetorical features (e.g., Charney & Carlson, 1995). For students in the Type A groups (both instructed and noninstructed), the Type A essays were labeled "effective Translantian essays" and the B essays "ineffective"; for students in the Type B groups, the B essays were labeled "effective" and the A essays "ineffective." All students were given 15 min to read these four essays (the slowest reader on the pilot test needed 15 min; all others finished considerably before this time). The target rhetorical features (signals of authorial presence, epistemic modality, cohesion strategy, reader engagement strategy, and organization) were highlighted, bolded, and/or underlined in the instructed students' essays. The noninstructed students received essays that had only been labeled "effective" or "ineffective," but no further marking was present.
After reading the four example essays, the students either (a) answered basic comprehension questions about the sample essays (e.g., Did the author of essay A1 want to live in the city pictured there? Why or why not?) (10 min) or (b) read a short synopsis and listened to a lecturette on the target rhetorical strategies and the reasons they were valued by the Translantians (10 min). For example, the instruction concerning the use of epistemic modality in the A essays involved telling students in the Type A group that the Translantians valued the use of forceful claims in persuasive essays, and for this reason, made liberal use of boosters. If a writer failed to use these words and phrases in a Translantian persuasive essay, her or his claims would appear hesitant and thus would be dismissed by the Translantians. Examples of boosters were also provided. The noninstructed students participated in (a) whereas the instructed students completed (b) (see Figure 1). At this stage of the treatment, all students were allowed to look back at the original essays and all students were invited to ask questions. However, at no point in time did the researcher mention any of the target rhetorical features to the noninstructed students or draw the noninstructed students' attention to those features.
Next, all students were asked to write a short persuasive essay to a native Translantian using the same "effective" style they had seen (Essay 1). All students responded to the same prompt (which asked students whether they wanted the government of Translantia to build a pictured city in their hometown or province and informed them that their audience was a native Translantian). Students were given 30 min to complete  Figure 1. Procedure this task, were not allowed to look back at the essays or summary sheets, and were videotaped as they wrote for the purposes of conducting a stimulated recall interview. 2 One week afterward, students were emailed a new prompt (which informed them that they had won a vacation to one of two pictured Translantian cities and that they needed to write a short persuasive essay to a native Translantian explaining which one they should be sent to; Essay 2). Students were instructed to work alone and to take 30 min. Finally, 1 month after treatment, students were emailed a quiz. On this quiz, the students saw an example of an "ineffective" Translantian essay (an essay that had little in common with either the Type A or B essays save for being three paragraphs long). Students were asked whether the essay conformed to the style of the "effective" essays they had seen 1 month prior (and why or why not). In addition, they were asked to rewrite any portion of the essay to make it more closely resemble the effective essays they had seen on Day 1. This procedure was chosen so as to allow students to explain their reasoning. The entire procedure is illustrated in Figure 1.

Data Analysis
Two broad measures were used to examine students' control of the target essay types. The first was a ratio, the total number of target rhetorical devices (appropriate to the type students had seen in the "effective" essays) divided by the total number of words. This ratio was calculated for both Essay 1 and Essay 2 for each student. Attempts at using the target rhetorical features, even if they were not grammatical, were counted. Interrater agreement with a research assistant on a random sample of 40 texts was 100% except for hedges and boosters (94% and 98%, respectively). Discrepancies were resolved through discussion.
The second measure was an adaptation of Henry and Roseberry's (1998) formula for determining how closely a student followed the target method of organization.
Move Score:

MP -0.5 (IM + MM) OM
In this formula, MP refers to the number of obligatory moves present in the student's text; IM is the number of "inappropriate moves" (here, moves other than description, opinion, or personal anecdote); MM is the number of misplaced moves; and OM is the number of obligatory moves in the genre (here, three for both Type A and Type B). As Henry and Roseberry (1998) note, "A text that has all the obligatory moves present in the correct order and contains no inappropriate moves will receive a score of 1. A text that contains none of the obligatory moves will receive a negative score" (p. 151). Interrater reliability on the random sample of 40 texts was 93% and discrepancies were resolved through discussion.
As noted earlier, students were emailed an example of an "ineffective" Translantian essay for the final quiz and were asked whether the essay followed the standards of the "effective" essays they had read 1 month earlier. For each rhetorical strategy (organization, reader engagement strategy, use of epistemic modality, authorial presence, and cohesion strategy) that the student either (a) correctly identified as being missing or otherwise ineffective or (b) rewrote so as to conform to the characteristics of the "effective" essays they had seen on Day 1, one point was awarded (thus, five points were possible).

Results
To examine differences in the participants' use of the target rhetorical devices on Essay 1 and Essay 2 and to examine the mediating roles of instruction (yes/no), group (native/ nonnative higher proficiency/nonnative lower proficiency), and essay type (A/B), a 2 × 3 × 2 factorial design with repeated measures was employed. The p value was set at .025 for all analyses. Initially, a completely saturated model (one containing all main and interaction effects) was fitted. Following recommendations in the literature (e.g., Crawley, 2002), the interaction terms were removed as none reached statistical significance (all p > .1 except for a three-way interaction for Time, Instruction, and Group at .09). A reduced model (one containing only main effects) was then refitted.
In the reduced model, there were significant main effects for Instruction, F(1, 113) = 80.210, p < .001, partial η 2 = .415; Type, F(1, 113) = 10.011, p = .002, partial η 2 = .081; and Group, F(1, 113) = 5.529, p = .005, partial η 2 = .089. Concerning Instruction and Type, the instructed students significantly outperformed their noninstructed counterparts, and the Type A students used significantly more of the target rhetorical devices compared with the Type B students. With respect to Group, post hoc tests revealed that the NS significantly outperformed the higher proficiency NNS (p = .007) and that the difference between the NS and lower proficiency NNS approached significance (p = .032). There were no significant differences between the higher and lower proficiency NNS (p = .885). These findings are illustrated in Figures 2 and 3.
An analogous procedure was used to examine the students' move scores on Essay 1 and Essay 2. As once again there were no significant interaction effects in the completely saturated model, these were removed, yielding a reduced model. In this reduced model, there were significant main effects for Instruction, F(1, 113) = 33.358, p < .001, partial η 2 = .229, and Group, F(1, 113) = 16.399, p < .001, partial η 2 = .225. There were no significant main effects for Time, F(1, 113) = .026, ns, or Type, F(1, 113) = .403, ns. Additional post hoc tests revealed that both the native speakers and higher proficiency NNS outperformed the lower proficiency NNS (p = .002 and p < .001, respectively). The difference between native speakers and higher proficiency NNS was not statistically significant (p = .059).

Discussion
The first research question asked if models and models combined with explicit instruction had a differential effect on students' production of the artificial essay types. Results indicated that the instructed students used significantly more of the target rhetorical features and followed the model organization more closely than those students who only received models (the noninstructed students). This was true for both Essay 1 and Essay 2 one week later. The instructed students also significantly outperformed the noninstructed students on the quiz 1 month after treatment.
In this following excerpt from an instructed higher proficiency nonnative speaker, the student, like the majority of the instructed students, used the entire range of rhetorical devices and employed the target method of organization.
(1)When the author was six-year-old, she went to New York with her parents. It is the city of skyscrapers and busy streets. The river runs through the city with grand bridges linking Manhattan and Brooklyn District. . . . Now we have City A and City B in the picture. City A may have a blue river runs across it. It will probably have bridges link the two parts of the cities. However, its skyscrapers might look like tall, but may not seem frightening. Therefore, people will probably feel comfortable living inside. On the other hand, City B might also have beautiful skyscrapers but it is possible they look threatening compared with small buildings around them. In addition, it probably does not have rivers run across it. As a result, people may not feel happy living there.
Please do not invite the author to City B! Invite her to City A! . . . Invite her to a city that will make her feel comfortable and happy.
In their stimulated recall interviews, the instructed students commonly made reference to their fictional audience and the efforts they made to write in the "Translantian" style. For these students, the models combined with the explicit instruction seemed to make the target features salient and accessible, and at the same time, gave them an understanding of why the features were valued by their fictional audience.
(2) Ishiro: "I was actually relieved to be able to be reading. I realized I was going to be writing the same type of essay, but I also understood why they wrote theirs, so . . . I knew what I wanted to write about, so that was exciting and I was just trying to, like the ideas were coming too fast, I was trying to slow my hand down, and remember that I'm not writing for an English speaker but for this Translantian." (3) Justin: "I'm trying to, just trying . . . follow the rhetorical conventions of what were the good samples with the kind of the style that they would appreciate, or that I think they would appreciate as Translantian readers." (4) Hana: "I try to be a person in the, I mean, I try to be a person who writes essay, uhm . . . and think about the audience who are in the Trans, Transia." Although the native speakers used more of the rhetorical features in comparison with the other groups, the effect of the instruction did not depend on the proficiency level of the participant (Research Question 2). Thus, it was not the case, for example, that only the higher proficiency students benefited from the instruction. This finding supports previous researchers' claims that instruction on rhetorical strategies need not wait until the student has reached an advanced level of proficiency (e.g., Ivanič & Camps, 2001;Johns et al., 2006).
It might be expected that because the target essay types were short (approximately 140 words), seeded with multiple examples of the target rhetorical features, juxtaposed with ineffective examples, and written on an accessible topic that the noninstructed students would have noticed and employed the target rhetorical features. However, this was not the case for the vast majority of noninstructed students in the present study, a finding that lends support to the results reported in Henry andRoseberry (1998), Smagorinsky (1992), and Stolarek (1994). The noninstructed students did comment that the essays were unusual; however, their stimulated recall data provided little evidence that the basis for this judgment was on the target rhetorical features. The noninstructed students mentioned a number of supposed characteristics of the model texts (e.g., use and placement of commas, type of adjectives, and length of sentences), but this may be taken as evidence of students extracting irrelevant details from the model texts (cf. Werner, 1989). Although lack of verbalization cannot be taken as isomorphic with lack of noticing (the lower proficiency students in particular may not have been able to explain to the researcher in English what they noticed about the texts), the fact that noninstructed students persistently focused on grammar, vocabulary, content, and "irrelevant details" may at least partially support the hypothesis that the target rhetorical strategies had not been the primary focus of their attention.
The reason for the noninstructed students' difficulties thus may lie partly in the salience of the target rhetorical features. Researchers have long noted that certain rhetorical strategies lack salience (e.g., epistemic modality; Hyland, 2000;Low, 1996), and without explicit instruction, students may fail to notice these strategies. The salience of the rhetorical features can also be undermined if the genre is unfamiliar. As Chu, Swaffar, and Charney (2002) noted, students' reading, processing, and recall of texts can be hampered if the text's method of organization or use of language is perceived as unfamiliar. Students' reading strategies may also be at work: If students are engaged in bottom-up decoding (reading word-to-word), this can undermine the noticing of rhetorical and organizational features (e.g., Hirvela, 2004;Leki, 1993).
However, cognitive factors are not the only reasons for the noninstructed students' difficulties with the target rhetorical features. Stimulated recall data provided evidence that some of the noninstructed students did in fact notice certain rhetorical features (the most common being the authorial presence strategies and cohesion), but not understanding why the features were employed in the texts (beyond the very general statement that the Translantians valued them), the students frequently chose to omit certain features because they conflicted with their preferred styles of writing.
(5) Fang: "Yeah, so I may have made the second, er the third half, the third part more passive due to the gratingness of not being consistent" (6) Melinda: "There were a few things that they did, like they started sentences with "and" and stuff like that . . . which I didn't do, but I was like, I was thinking that would be acceptable, but I still didn't put in there . . . I think it was just not, it was unnatural for me." Although this type of resistance is to be expected with an artificial essay type, it has also been well documented in studies of L2 learners' encounters with authentic genres (e.g., Cadman, 1997;Fox, 1994;Ivanič, 1998;Petrić, 2005). This conflict between the "new style" and students' preferred or previously learned ways of writing also appeared to make the task difficult for the noninstructed students in the present study: (7) Kwan: "Because it's different from the essay I used to write, so it's even more harder for me to write, yeah." (8) Jiao: "Difficult for me because it's not the writing style that I use right now." (9) Yuuta: "It was, it was quite . . . difficult in a sense that it, you know, by reading the good essays and the bad essays, sort of trying to figure out what constitutes as a good essay and what constitutes as a bad essay, and, it's quite different from like let's say a north American academic English would consider to be good writing." (10) Hoshi: "I think it's really hard because I . . . I get used to writing in English so, maybe there's a habit going on when I write." The instructed students found the Translantian style "odd" as well, but having a rudimentary understanding of the sociorhetorical reasons underlying the use of the conventions, they were able to temporarily set aside their preferred styles of writing and employ those expected by their audience.
There were, however, a handful of exceptions to this general trend-namely, noninstructed students who were able to glean all of the target rhetorical features from the model texts and use them in their writing, without having had instruction. For example, Maggie, a noninstructed native speaker of English, used all of the rhetorical strategies associated with the B essays (third person pronouns and phrases referring to the author, hedges, linking adverbials associated with writing, and directives) in Excerpt (11) below: (11) The author of this essay once went to Seattle when she was a child. Seattle had many entertainment areas and also many eating establishments. Furthermore, Seattle was a city of culture and refinement.
The author of this essay feels that the picture of the suggested city is like Seattle. It shows promise of having many places to eat, recreate and seek culture. . . .
The author of this essay urges the Translantian government to aid in the construction of such a city. Do not be distracted by naysayers or politicians who suggest not to build. The building must begin now.
However, the only students who were able to do so were a small number of native speakers who had indicated on their questionnaires that they were avid readers and writers. Presumably these students, due to their previous experience reading and writing a wide variety of genres, were familiar with the typical dimensions of difference across genres and possessed the analytical tools they needed to make sense of the lexicogrammatical and organizational patterns they saw in the new texts (cf. Bereiter & Scardamalia, 1984). None of the noninstructed nonnative speakers were able to reproduce all of the features of the model texts, even those at the higher level of proficiency. This result supports claims made previously in the literature that inductive approaches may be insufficient for L2 students and that explicit instruction can facilitate students' noticing and use of target rhetorical features (e.g., Hyland, 2003).

Conclusion
No claim is being made here that explicit instruction, and especially the form employed in the present study, is sufficient. To move from genre competence to genre performance, students will clearly need more instruction, writing opportunities, feedback, and immersion-based experiences. Nor can it be concluded from this study that the noninstructed students would not have been able to pick up the target rhetorical features eventually if they had received more exposure to the models. The most serious limitation, the use of an artificial essay types, must also be borne in mind. Authentic genres are, by definition, "socially situated and culturally embedded" (Bawarshi & Reiff, 2010, p. 197). Artificial essay types are "de-situated," with the audience and communicative purpose reduced to the barest of fictions. This raises serious questions about the external validity of the results, making replication studies employing authentic genres and finer-grained measures of genre learning a necessity.
Despite these limitations, the results of the present study do suggest that models are insufficient for all but the most talented writers. Just as models do not illuminate the writing processes that gave rise to the final text, they do not reveal the sociorhetorical reasons underlying the use of particular lexicogrammatical or organizational strategies. Without this information, students may be reluctant to set aside preferred or previously learned ways of writing. Models alone may help students gain "some procedural knowledge" (Tardy, 2006, p. 92) or a "dimly felt sense" (Freedman, 1993, p. 230) of how the genre works, but if we are to facilitate students' noticing, and crucially, understanding of the target rhetorical strategies-and in the process, help students build a solid foundation for later more authentic encounters with the genre-then we need to recognize the importance of explicit instructional techniques.

Sample Translantian Essays
Prompt: Would you want to live in the city pictured below? [picture of futuristic city] Type A. This is a city of steel and a metropolis of metal.
It catches your eye with sharp corners and steep walls. Also, it catches your attention with colorful signs and loud sounds. Who wouldn't want to live here? Who wouldn't want to make this their home? Who wouldn't want to call this their city?
When I was five, my family traveled to London. We were impressed by the grand buildings and awed by the massive concrete fixtures. Besides, we loved taking the metro to sites around the city. And we definitely loved riding the Double Decker buses.
I would definitely like to live in this city. It is beyond doubt the greatest city on earth. Certainly, life would be enjoyable in this city, just like London. Also, my life doubtlessly would be very efficient. So, I would really like to live here.

(continued)
Type B. Ten years ago, the author of this essay had the opportunity to travel to Tokyo, Japan. In that city, the author could see a wide variety of buildings, skyscrapers, parks and roads. Moreover, he was impressed by the efficiency of the city. The city ran like a well-oiled machine. Therefore, the author promised someday to return to this city.
The city pictured above, however, probably would not be very efficient. Due to its shape and location, the city most likely would suffer from heavy traffic and could have problems with crime. In all likelihood, the city would be a depressing place to live. In addition, it might even have problems with pollution.
Do not let the author live in such a place. Let the author stay in his home city. Furthermore, do not make the author move to a place this depressing.

Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author received no financial support for the research and/or authorship of this article.

Notes
1. Campus-wide, the Writing Proficiency Examination (WPE) passing rate is 96% for native speakers of English and 61% for nonnative speakers of English. In this study, the means and standard deviations of the WPE scores for the three groups were as follows: lower proficiency NNS: M = 9.15, SD = .67; higher proficiency NNS: M = 12.09, SD = .75; NS: M = 12.54, SD = 1.21. None of the native speakers in the present study had received a nonpassing score on the WPE and none were enrolled in remedial writing courses. 2. A stimulated recall interview employs visual cues from the original task to elicit detailed information from the participants on their thought processes and attentional foci at the time of the original task (Gass & Mackey, 2000). The purpose of this procedure was to gather information on what students noticed in the sample texts. Students were videotaped individually and were interviewed on a one-on-one basis immediately after Essay 1.