Pergamon Computers in Human Behavior, Vol. 10, No. 4. pp. 5 1l-527, 1994 Copyright Q 1994 Elsevier Science Ltd Printed in the USA. All rights reserved 0747-x32/94 $6.00 + 00 Writing as Process and Product: The Impact of Tool, Genre, Audience Knowledge, and Writer Expertise Florida At/an tic University CmMichael Levy Universj~ of Florida Abstract - Two experiments investigated the impact of writing tool (word processing or handwriting), genre (narrative or exposition), and audience Cfamiliar or unfamiliar) on measures of writing quality, syntactic complexity, and number and type of initial text production revisions. In the first, 84 undergraduates with little word processing experience wrote letters by hand or computer. The 64 subjects in Experiment 2 were experienced college writers who always wrote by computer. Subjects composed more syntactically complex letters of higher rated quality to an unfamiliar audience than to a familiar one. ~andwri~en letters were of higher rated q~li~ than word processed. Although there were more total revisions when using a word processor, there were more text-preserving than meanin~ul revisions. The number and distribution of revisions also depended upon the writers’ level of experience. The Hayes and Flower (1980) model of the writing process remains a useful heuristic, but our data indicate that it warrants extension. Requests for reprints should be addressed to Dr. Sarah E. Ransdell, 2912 College Ave., Florida Atlantic University, Davie, FL 33314. E-mail: RANSDELL@FAUVAX 511 512 Ransdell and Levy Research on the psychology of writing has ranged from studies primarily interested in analysis of the written product to those whose goal is to explain the writing process as it unfolds through time. Verbal protocol analysis, pioneered by Hayes and Flower (1980), was one of the first process-tracing methods. Evidence emerging from protocol analysis suggests that strategic knowledge plays a major role in all phases of writing. Different kinds of strategic knowledge can influence the written product, but the mechanisms by which such knowledge operate are far from clear. Given the complexity of adult composition, strategic knowledge probably does not combine in simple, additive ways. For example, the information stored about audience knowledge may be organized quite differently from knowledge about rhetorical genres, such as how to write a narrative or persuasive letter. Some of what a writer knows about an audience may be only basic demographic information. Other information about an audience, such as what they know about the topic of a composition, may need to be inferred. Thus, the process of writing results in a dynamic interaction among the writer’s knowledge representations. Just as it is possible that strategic knowledge structures can interact during the writing process, it is possible that these structures can influence the products of writing in complex ways that the study of individual structures, no matter how rigorous, will never be able to detect. Thus, studies that focus only on the contribution of a single variable, such as audience knowledge, can provide only narrow conclusions about either the writing process or the written product. The present research investigates the role of several kinds of knowledge structures as well as writing tool on both writing process and product. Our goal in this research is to examine how the strategies that result from writers’ access of their structures affect multiple attributes (e.g., quality, syntactic complexity) of compositions, as well as the kinds of revisions that are created during text production. We also aim to assess, using a uniform experimental methodology, the relative contributions and interactions of these variables that have been studied in virtual isolation from one another in research that has little methodological overlap as shown in Table 1. Most experimental research has focused on the impact of writing assignment and audience knowledge in terms of measures such as syntactic complexity (Crowhurst & Piche, 1979; Hunt, 1983; Langer, 1984), cognitive effort or capacity (Kellogg, 1987; Reed, Burton, & Kelly, 1985), draft to draft revisions (Bean, 1983; Bridwell, 1980; Collier, 1983; Lutz, 1987), and writing quality (Rubin & Rafoth, 1986; Witte & Faigley, 1981). Process-oriented research has been mainly restricted to protocol analysis (e.g., Hayes & Flower, 1980) and to cognitive effort studies (Kellogg, 1987; Reed, Burton, & Kelly, 1985). Very few studies have focussed specifically on the relationship between process and product measures in order to explain the complexity of written language production. The methodology that we use showcases word processed writing because it is particularly amenable to analysis of both process and product. The increasing use of the word processor as a writing tool calls for steps to determine its influence on written language. In a review of recent research, Cochran-Smith (1991) concluded that the effects of word processing clearly interact with preexisting skills and strategic knowledge. While there have been reports of systematic differences between word processed and handwritten text (Collier, 1983; Joram, Woodruff, Bryson & Lindsay, 1992; Lutz, 1987), the few experimental studies that exist focus on either the cognitive processes or the products themselves, but usually not both. Process and product 513 Table 1. Cited References by Focus, Method, Variables Studied, and Type of Subjects _IFocus Method Product Ohs. Expt. independent Var. Dependent Var. Length t-units Clause length Errors Quality 5 high-knowledge, 5 low-knowledge adults Witte & Faigley (1981) Topic knowledge Clause length Quality 99 10th grade students Langer (1984) Audience’ Genre2 Age3 Clause length12,3 t-units2t 120 6th grade students, 120 10th grade students ~rowhu~t PI Piche (1979) Expertise’ Genre9 Cognitive en~gement~ 12 Clause lengthfp2 t-units’12 Quality1 p2 21 low-, 21 average-, 21 high-knowledge first-year undergraduates Reed et al. (1985) 35 students in first-year Rubin & college composition course Rafotth (1986) Expertise’ Genre2 Time (draft)3 ~uaii~1y2~3 16 low-, 16 average, 16 Anxietyf*213 high-ability Education Apprehension’ ,293 students with no word processing experience Reed (1992) Not applicable Testimonials 4 first-year college students with no word processing experience Bean (1983) Draft1 Revisions Quality’ Length’ 100 12th-grade students Bridwell (1980) Topic knowledge’ Processing time Cognitive effort1 30 high-knowledge, 30 low- Kellogg (1987) knowledge undergraduate Tool Style Time (session) Quality Length Self-reports 8 professional writers (graduate TAs) with word processing experience BridwellBowles et al. (1987) Dbs. Tool’ Revisions1 Quality 4 college composition students with no word processing experience Collier (1983) Expt. Expertise1 Tool2 Total time2 Revisions2 Length2 4 professional writers, 3 experienced writers (graduate TAs) with word processing experfence Lutz (I 987) Tooll, Topic* Revision’ ,2 Quality2 20 advanced first-year college students Hawisher (1987) Audience’ Revisions1 A~itude scores1 87 undergraduates Redd-Boyd & Slater (1989) Ohs. Expt. Product and Process Researchers Expertise Expertise’ Genre2 Social cognitive abilitys Process Subjects (Tsbe I continued~ 514 Ransdell and Levy Table 1. Continued Focus Method Independent Var. Dependent Audience awareness adaptation Tool’ Time of revision* Instructions prior to 2nd draft3 Skill of subjects4 Var. Subjects and Quality3s4 Creativity Nature of revisions Various measures from thinking aloud protocols 31 average and above average 8th grade students Note. Superscripted numerals indicate where the researcher(s) relationships between the variables sharing the same numeral. WORD PROCESSING, Researchers REVISION reported Joram et al. (1992) statistically significant AND QUALITY The most widely observed effect of word processing on writing is that it results in greater amounts of revising (Collier, 1983; Lutz, 1987). However, the correlation between absolute number of revisions and writing quality is not always strong, and so many researchers categorize revisions into those that change the meaning of the text and those that do not (e.g., Faigley & Witte, 1984). Several researchers have reported that word processed writing tends to contain a higher proportion of textpreserving revisions to revisions that actually change the content or meaning of the text (Bridwell-Bowles et al., 1987; Lutz, 1987). Hawisher (1987) observed a negative correlation between the number of text-preserving revisions and improvement in writing quality in first to final drafts created by word processing. She also found a positive relationship between the amount of meaningful revisions and writing quality (meaningful and text-preserving changes do not always “trade off”). These reports must be interpreted cautiously because a design decision limits their generalizability: Revision was measured by changes from one completed draft to the next. A more fine-grained process approach would look at revisions at the point of initial text production. Only with this latter approach is it possible to categorize the complete set of revisions that occur. In fact, the relative ease with which writers can make point-of-entry modifications in word choice, order, and organization is perhaps a large part of what makes composing with word processing unique. The present research is concerned with the process of revision as it occurs within a draft. We use a special purpose terminate-and-stay-resident (TSR) program (Ransdell, 1990) to record when each keystroke is made and later replay the composition in real time. The technique enables us to study various types of text and idea manipulations at the point of utterance. This approach has been used by Bridwell-Bowles and her associates (Bridwell-Bowles, Johnson, & Brehe, 1987) but has been limited to case study procedures and to experienced computer users (but see Bonk & Reynolds, 1992 for a study of an adolescent population). In a cognitively challenging task environment, when the mental resources needed to perform a subtask (such as revising) are reduced, the frequency with which that subtask occurs may increase. For example, many have observed that keystroking errors and minor surface changes are much easier to accomplish with a word processor than on typewriter. Accordingly, we anticipated that our moderate- Process and product 515 ly proficient touch typists would revise more when the properties of their writing tool facilitated rather than hindered revisions. Specifically, we predicted that writers would exhibit more revising when they composed using a word processor than when handw~ting, but a smaller prounion of such revisions would change the meaning of the text. AUDIENCE AND SYNTACTIC COMPLEXITY Constraints on idea and text production can also be imposed upon writers by altering audience knowledge. Crowhurst and Piche (1979), Redd-Boyd and Slater (19891, Reed, Burton and Kelly (1983, and Rubin and Rafoth (1986) have reported how audience for a composition can increase syntactic complexity and improve quality. Very few past studies have simultaneously compared product indices to process measures such as the nature and qu~tity of revisions. Our approach examines the impact of audience familiarity on writing quality, syntactic complexity, and type and amount of revision. Assigning a specific audience to writer’s essays has been found to increase motivation, the use of audience-based strategies, and actually improve the writing’s persuasiveness (Redd-Boyd & Slater, 1989). Furthermore, when writing in a persuasive genre, writers are most likely to benefit from knowledge of audience because they can tailor their arguments to appeal to what they know about the reader (Crowhurst & Piche, 1979; Rubin & Rafoth, 1986). Skilled writers are often more aware of audience than less skilled (Hayes 2% Flower, 1980; Rubin & Rafotb, 1986). Assigning a distant, unfamiliar, or adult audience frequently leads writers to produce syntactically more complex documents as revealed by longer average clause length (Crowhu~t & Piche, 1979). This may reflect an impression-management function - to put one’s best foot forward - when writing to authority figures (Kirsch, 1991). We predict that our writers will create letters of greater quality and syntactic complexity when writing to an unfamiliar audience than to someone they know well. From the writer’s perspective, the point of revising is to produce a better document. It is an open empirical question, however, whether there is a strong positive correlation between the extent of meaningful revisions and the rated quality of a finished written product. To the extent that such a correlation generally exists and to the extent that writers are motivated somehow to compose at their best for an unfa~liar audience, we should find a relationship between audience and the quantity of meaningful revisions. This relationship may hold for revisions made during the creation of a single draft of a document or between two drafts of the same document, or both. GENRE OF DISCOURSE A third factor suggested by the Hayes and Flower model that may influence writing is the genre requirements of the assignment. Reed, Burton and Kelly (1985) asked students to write in one of three genres (descriptive, narrative, or persuasive) while performing a secondary task, responding to a tone, which served as a measure of cognitive engagement. Writing in the narrative genre produced the fastest reaction times to the tone, indicating the least amount of engagement. Subsequently, 516 Ransdell and Levy Reed (1992) found that both high- and low-ability writers produced their best essays when writing narratives and their worst when writing persuasive essays. These results corroborate a finding in Ransdell’s (1989) study where 86% of the writers described persuasive writing as being more difficult than narrative. In addition to simple effects, interactions between genre and audience familiarity have also been reported. Crowhurst and Piche (1979) found significantly longer average clause length in argumentative essays when the intended audience was less familiar to the writer (i.e., a teacher) than when the audience was very familiar (i.e., a best friend). In contrast, in narrative writing, clause length was not affected by audience. Crowhurst and Piche suggest that persuasive writing demands greater attention to audience than narrative or description. All of these findings are also consistent with an explanation that is based upon the writer’s knowledge of the genre. Because knowledge is commonly considered a continuous rather than a binary entity, these data may reflect the fact that the extent of a typical subject’s knowledge of narrative production rules are more extensive, more practiced, and more automatically invoked than the rules for creating persuasive documents. Either conceptualization, however, leads to the same set of conclusions: writing in a narrative genre should result in more meaningful revisions and higher rated quality than writing in the persuasive genre. In summary, we predict that measures of quality, revision, and syntactic complexity will vary as a function of writing tool, genre, and audience familiarity. Hayes and Flower (1980) described a process model that makes no explicit predictions about written products. An extension of this theory suggests that when writers have access to strategic knowledge, they will bring that knowledge to bear in composing higher quality documents. Thus, we explicitly manipulated the subjects’ knowledge of their audience, the writing tool, and genre. In Experiment 1 we vary tool, audience familiarity, and genre between subjects. In Experiment 2 we examine the effects of the writers’ expertise in terms of college writing experience in combination with the effects of audience and genre knowledge. EXPERIMENT 1 Method Subjects. The subjects were 84 introductory psychology students1 at the University of Maine who voluntarily participated for extra credit in their class. Experience with word processing, typing ability, and history of college writing courses were collected in an initial class period. Subjects were invited to participate only if they had little or no word processing experience, but fairly good self-reported typing ability. Sixty-eight percent of the subjects had written at least one paper for a college assignment, 67% were taking a freshmen composition course at the time, 59% reported writing at least two personal letters a month, and 91% considered themselves to be at least average writers compared to their college peers. Only seven subjects had ever used any word processing software that they could name, and none had used the program used in this study. Design. The design was 2 (writing tool: handwriting or word processing) by 2 (audience familiarity: familiar or unfamiliar) by 2 (genre: narrative or exposition) Process and product 517 between-subjects factorial experiment. Holistic quality ratings, syntactic complexity as measured by mean clause length, and total number and type of point of text produc~on revisions served as dependent measures. A~~arQt~s. Some subjects wrote using a popular word processing program on IBM-PC compatible microcomputers, The word processor was used in a simple and generic form; that is, the screen was initially blank and the subjects were informed about only an extremely limited subset of editing cormnands on a colorcoded template. The other subjects wrote on notebook paper using indelible pens. Reed (1992) has noted the difficulty in directly comparing word processing with handwriting. Note that several characteristics of our procedure for collection, analysis, and interpretation attempt to take this inherent difficulty into account. Procedure. All students were first trained in some of the basics of the word processing software, including how to insert or delete text and how to save their document. Half of the subjects wrote by hand using the pen and paper provided and half used the computers and the word processing software. Those writing by hand were asked to make any changes or revisions visible by crossing them out with a single line. All subjects wrote for 20 min; they were warned when only 5 min remained. One of our aims was to devise a paradigm in which the writer’s knowledge of the topic and of the genre would not limit the production of prose. We, therefore, selected a topic about which students at most universities would have had direct experience over an extended period of time: college course selection and registration. Half of the subjects wrote a letter to a family member or close friend; the remainder wrote a letter to the president of the University of Florida. All subjects were told they would be given copies of their letters to send in order to enco~age them to be as realistic as possible. To protect their privacy, subjects were told not to sign their names to the letters. Half of all subjects wrote a descriptive narrative. They were asked to describe the problems that they had encountered in selecting and registering for classes at the University of Maine. In addition to describing the process, they were asked to include what they actually did and how they felt about their personal experience. The other subjects were asked to write an argumentative exposition. They were asked to argue for some innovative ways that the university could improve how students select and register for classes. They were asked to state as many arguments as possible to support their ideas. Cards with genre and audience instructions were placed in front of each subject’s work space. All h~dw~tten letters were transcribed into word processed form so raters could be blind as to tool. Two judges independently rated each letter on 13 dimensions of writing quality based upon a tool developed in holistic quality assessment at the University of Maine as part of English placement examinations (Nees-Hatlen, 1989; see Appendix A). Training sessions for the raters followed closely the procedures from those of the placement exams so as to maximize interrater reliability. Each judge rated each letter one dimension at a time. Overall interrater reliability was r = .84 for total quality scores across the four raters. A total quality score was calculated for each subject by averaging the 13 evaluations of each letter across both judges, then converting these to percentages. This interrater reliability is consistent with previous research (see Cooper, 1977). The analysis of point of text production revisions measured in the present study foilowed a commonly used taxonomy of revision types (Faigley & Witte, 1984). 518 Ransdell and Levy The main distinction was whether or not a revision changed the meaning of the text. Within the bounds of meaningful changes are those that reflect changes in the topics or concepts that are included in the essay as well as those that change the general organization or outline of the text. Text-preserving changes include both surface changes such as those of spelling, tense, and punctuation, as well as additions, deletions and substitutions that preserve meaning. Results: Experiment 1 writing tool. Separate 2 x 2 x 2 between-subjects ANOVAs were conducted on each of the dependent measures. The first revealed a main effect for writing tool; handwritten letters were judged to have higher holistic quality (mean score = 7 1%) than word processed (mean = 65%) [F(l) 68) = 6.65, p c .Ol]. No interactions were significant with tool for quality scores (see Table 2). Syntactic complexity (as measured by mean clause length) was correlated with quality scores (Pearson r = .32, p < .004). This significant correlation found between the two product measures was the only such relationship among the four dependent measures. Syntactic complexity was not, however, influenced by writing too1 [F< 11. Subjects made significantly more total revisions within word processed letters (mean = 50) than for handwritten (mean = 6.7) [F(l, 68) = 162.6, p c .OOOl]. However, after removing the types of revisions that did not influence meaning, such as typos and spelling changes, handwritten letters (mean = 6.7) actually contained more revisions than did word processed letters (mean = 4. l), [F(l, 68) = 8.88, p < .004]. The assigned genre had no reliable influence on any of the four dependent measures. Holistic quality was affected by audience familiarity, with letters to an unfamiliar audience receiving a higher score (71%) than to a familiar audience (65%) [F(l, 68) = 5.85, p < .Ol]. Letters written to an unfamiliar audience contained significantly more words per clause (mean = 8.3) than letters to a familiar audience (mean = 6.8), [F(l, 68) = Audience. Table 2. Mean Quality Scores, Clause Length, and Point-of-Utterance Tool, Genre, and Audience in Experiment 1 Revisions By Revisions(#) Quality(%) H Tool Genre Aud WP Clause Length (wrdskl) H WP (All) H WP (Mnfl) H WP 71 > 65 7.5 = 7.5 6.7 < 50 6.7 > 4.1 N N N N E E 68 = 68 7.5 = 7.6 U U F 71 >65 F 8.3 > 6.8 E 27.1 = 29.3 U F 29.9 = 26.8 E 4.5 = 6.2 U F 6.9 > 4.0 Note. The following abbreviations are used in the table: H = Handwritten, WP = Word processed; N = Narrative genre, E = Expository genre; U = Unfamiliar audience, F = Familiar audience; and Mnfl = Meaningful revisions only. Process and product 519 .OOOl].Furthermore, in analyses involving all revisions, audience was not a significant effect [F < 11. In an ANOVA containing only meaningful changes in word processed letters, audience was a reliable factor. An average of 6.9 revisions occurred in letters to an unfamiliar audience versus 4.0 to a familiar [F( 1,76) = 9.64, p c .003]. Unlike clause length, the total number of revisions was unrelated to overall quality, Pearson Y = -. 18, p = .l, nor was meaningful only related to quality, r = -.02. Neither revision measure was related to clause length nor were the two measures of revision correlated with one another. A significant interaction occurred between audience and tool in the analysis with meaningful revisions [F(l, 76) = 4.05, p < .04]. Handwritten letters to an unfamiliar audience contained relatively more meaningful revisions than word processed letters to a familiar audience. The analysis of total revisions produced a significant three-way interaction between audience, genre and tool [F(l, 76) = 4.64, p < .03]. The most striking pattern was that handwritten, exposition letters to an unfamiliar audience contained fewer total revisions than word processed narrative letters to a familiar audience. 21.10, p < Summary and Discussion of Experiment 1 A summary of the results shows that subjects composed more syntactically complex letters to an unfamiliar audience than to a familiar one. While the writers revised more often when using a word processor than when writing by hand, they made more text-preserving than meaningful revisions. Holistic quality was positively correlated to syntactic complexity. No other measures, however, were significantly correlated. Some dependent measures showed no effect of writing tool. Syntactic complexity was a powerful result of manipulating audience familiarity but was not affected at all by writing tool. Total number of revisions increased with word processing, but our real-time record allowed us to see that the bulk of these changes were text preserving. Other studies looking at draft-to-draft revisions have suggested this result (Collier, 1983; Lutz, 1987). Hunt (1983) has suggested mean clause length is the most parsimonious way to measure syntactic maturity or complexity, particularly for high school and collegeage writers. As writers mature, they tend to consolidate ideas into larger clauses rather than simply into longer sentences. Our results are, thus, similar in magnitude to those Hunt reported for 12th grade students. EXPERIMENT 2 In this experiment, we examine the effects of writer expertise in combination with the effects of audience and genre. Because there are many reports of differential cognitive processing by experts and novices in domains varying from encoding in short-term memory (Chase & Ericsson, 1982), problem-solving (Chi, Glaser, & Rees, 1982), decision-making (Northcraft & Neale, 1987), and reading (Daneman, Just & Carpenter, 1982) differences between the writing processes and products of experts and novices is anticipated. Studies that make such comparisons, however, often compare college faculty and professional writers to undergraduates, seldom recognizing that such groups differ in several noteworthy dimensions that might influence the writing quality in a specific domain. 520 Ransdell and Levy In the present study, we targeted individuals who had equivalent verbal SAT test scores, were from the same cohort, but who differed on dimensions that conceivably could influence their performance and their products. Specifically, we focussed on a group of students who had taken multiple college courses that required extensive writing and who often wrote letters in their free time, and compared them to a group who had deferred completing the university requirement for these classes with heavy writing demands and who seldom wrote letters in their free time. Method Nearly 1,000 students enrolled in a general psychology course at the University of Florida were given a survey to determine the extent of their experience with word processing programs, their typing speed, the college-level courses they had taken that required extensive writing2 and the number of personal and business letters written each month. Those people who had some prior experience using a word processing package and who stated that they could touch type at least 10 words per min were included in the candidate pool. Those invited to participate met one of the following criteria: (1) they had taken no more than one course requiring extensive writing and wrote an average of no more than one business and personal letter per month, or (2) they had taken two or more courses requiring extensive writing and wrote an average of four or more business and personal letters per month. Those who met the first criterion are referred to as “Low Expertise” and those who met the second are identified as “High Expertise.” Usable data were collected from 64 subjects; data from six other subjects were voided owing to equipment malfunctioning and procedural errors. Subjects were tested in groups of two to six. Subjects. Materials and Procedure Because they all had prior word processing experience, it was necessary only to familiarize subjects with the keyboard layout of the IBM PS/2 computers and to give them practice in typing and editing a sample paragraph. Half of the subjects were then asked to write to a close friend or relative (familiar audience) about selecting and registering for classes at the University of Florida. The remaining subjects wrote on the same topic, but addressed their letter to the President the University of Maine, an unfamiliar audience. Within each audience group, half of the subjects wrote a narrative letter describing their experiences and their feelings; half wrote an expository letter arguing for changes and improvements. Low-expertise and high-expertise writers were equally represented in each of these four groups. There were eight subjects initially in each of the 2 x 2 x 2 (Audience x Genre x Expertise) cells of the experiment. Immediately after being given their writing assignment, the TSR program was started to enable monitoring and recording of the writer’s keystrokes. When the subject had written for 20 min, the researcher gave each subject a printed copy of their first draft to refer to as they tried to “strengthen and improve the quality of their letter” by using the editing features of the program to create a second draft. The subjects spent the final 15 min editing their documents on the screen. The quality rating procedure was the same as in Experiment 1. Two of the three judges had evaluated the letters composed in the earlier experiment; the assessments Process and product 521 of a third trained judge were included to resolve differences in ratings between the first two judges. A second pair of judges determined the length of each clause written, using the same criteria reported in the earlier experiment. Five additional independent judges initially assigned each letter a value along a 7-point Likert scale indicating whether it was primarily narrative or expository or some mixture of the two genre. Several weeks later they reread each letter and used another 7-point scale to record their estimate of the writer’s expertise; they were to focus on the quality of the exposition and the structure of the document, ignoring typing errors. Analyses were conducted on median ratings of judged genre and expertise. Results: Experiment 2 Level of expertise. Subjects initially categorized as low in expertise reported writing an average of 1.2 business and personal letters per month and had taken an average of 0.6 college courses that required substantial (6000-word minimum) amounts of writing. In contrast, subjects categorized as high in expertise reported writing an average of 3.7 letters per month and having taken an average of 2.3 courses requiring substantial writing. The high- and low-expertise subjects were reliably different on these two dimensions (p < .OOl), but did not differ in reported verbal SAT or ACT scores or reported typing speed. Our judges’ independent assessment of the writers’ expertise provided a measure of the validity of the preexperimental assignment to high- and low-expertise groups. Subjects assigned to the low-expertise group produced letters that were judged as being written by poorer writers than those assigned to the high-expertise group (means = 3.8 vs. 4.7, along a 7-point scale with 1 = poor). While this effect was significant [F(l, 114) = 10.24), p c .002], it is important to note that judges did not evaluate the writers as either extraordinarily good nor poor. Assignment to high or low expertise corroborated the quality assessment results. The letters written by high-expertise subjects were rated as superior in quality to those written by low-expertise subjects [F(l, 122) = 5.83, p < .02]. A significant expertise x draft interaction [F(l, 122) = 10.7, p c .OOl] revealed that the nearly 5point superiority in overall quality for the letters written by the high-expertise subjects on the first draft was reduced to about 4 points for the second draft. The average clause length was longer in the letters written by the high-expertise subjects [F(l, 56) = 4.8, p < .02], but only in the first drafts. Audience. Audience was a powerful variable in this experiment, influencing not only the syntactic complexity of the subjects’ letters, but their length, and the number of revisions. Letters to unfamiliar audiences contained significantly longer clauses than those to familiar audiences [F(l, 122) = 22.7, p < .OOl]. Letters addressed to a friend or relative were also reliably longer than those addressed to a university president [means = 370 vs. 314 words, F(1, 122) = 12.1, p < .OOl]. This significant relationship occurred in the first drafts and increased when subjects prepared their second drafts. The revised letters to friends and relatives were 21% longer than the original versions, but the second drafts to the university presidents were only 11% longer than the initial drafts, producing a significant audience x draft interaction [F(l, 122) = 8.1,~ < .Ol]. Genre. As in the previous experiment, the effects of genre on the process and products of this writing assignment were small. For example, expository letters 522 Ransdell and Levy contained longer clauses than narrative letters and they were also judged as higher in quality than narrative letters, but these differences were not reliable. Inspecting the letters, we saw few strongly presented arguments in the expository letters. Therefore, as noted above, we asked a new set of independent raters to read each letter and judge whether it was primarily narrative, primarily expository, or contained various amounts of each genre. Subjects were successful in writing to the task assigned when they used the narrative genre (mean judged genre = 1.4, where 1 represents a primarily narrative letter containing no argumentative, expository elements). In contrast, they were not particularly successful in producing purely expository prose. The mean judged genre was 4.5, indicating that while there were expository elements present, they occurred only slightly more frequently than narrative elements. The difference between these judged genre means was highly significant [F(l, 114) = 104.0,~ < .OOl]. Independent of the genre that they were assigned, subjects tended to fashion letters containing more narrative elements when they were writing to a familiar audience than to an unfamiliar audience ljudged genre means = 2.6 vs. 3.3, F( 1, 114) = 6.5, p c .Ol]. There were strong social pressures to write something in this very public setting. The subjects seemed to resort to using the more familiar, well-used, and accessible schema for composing narrative prose rather than to write nothing at all. The four-way interaction [F(l, 40) = 7.19, p < .Ol], between audience, genre, expertise, and type of revision was the only significant interaction, In general, it indicated a pattern of more meaningful revisions by the high-expertise subjects when they wrote the expository letters written to an unfamiliar audience than by low-expertise subjects (who made about the same number of meaningful revisions regardless of audience and genre). This relationship was almost completely reversed in the pattern of text-preserving revisions. In that case, the high expertise subjects made a relatively small number of revisions (regardless of genre and audience) but low expertise varied greatly across the combinations of audience and genre. Revisions to the second draft were assessed by comparing the first and second drafts word by word. In this analysis, a “meaningful revision” included adding or deleting a single word or adding or deleting a multiword phrase that constituted a clause. Overall, subjects made an average of 1.3 text-preserving changes to the second draft, but an average of 11.9 meaningful changes [t(59) = 13.9, p < .OOl]. Neither the number of meaningful nor the number of text-preserving changes were reliably influenced by genre, audience, or expertise. Summary and Discussion of Experiment 2 Writer expertise had some significant effects on quality, revision, and clause length. “High” expertise subjects’ letters were rated as higher in quality and contained a larger percentage of meaningful revision relative to text preserving. Audience familiarity remained a robust factor influencing clause length but the “high” expertise subjects were less affected by this variable suggesting that they generally use strategic knowledge about audience in their written work. The four-way interaction between audience, genre, expertise, and type of revision in Experiment 2 was the only reliable “higher-order” interaction. In general, it showed that there were more meaningful revisions by the high expertise subjects when they wrote the expository letters written to an unfamiliar audience than by low expertise subjects. The low expertise subjects made about the same number of Process and product 523 meaningful revisions, regardless of audience and genre. This pattern was almost completely reversed when considering text-preserving revisions only. In this case, the high-expertise subjects made a relatively small number of revisions (regardless of genre and audience) but low expertise varied greatly across the combinations of audience and genre. This result was predicted on the basis of several studies indicating that less skilled writers spend relatively more time manipulating text than they do creating ideas (Bean, 1983; Bridwell, 1980; Hayes & Flowers, 1980; Hillocks, 1986; Sommers, 1979). GENERAL DISCUSSION A major strength of the present study is that several results from previous research have been replicated in two well-controlled experiments. Writing to an unfamiliar audience improved writing quality, increased syntactic complexity of compositions, and led to a higher proportion of meaningful revisions than when writing to a familiar audience. High-expertise college writers wrote compositions of higher quality and greater syntactic complexity than low-expertise writers. The compositions of high-expertise writers also differed from low by containing a higher number of meaningful revisions, especially when writing expositions and when writing to an unfamiliar audience. Added to these results suggested by previous research are the important effects of writing tool. Word processed writing contained more total point-of-utterance revisions, but was of poorer quality and contained fewer meaningful revisions than handwritten writing. The Hayes and Flower (1980) model remains a useful heuristic, but our experimental data indicate that it warrants extension. On the basis of our findings, we propose that writing tool should be included as an important factor in models of the composition process and that the relationship between product and process measures should be considered. For example, it is important to note that the dependent measures, (syntactic complexity, quality, and amount and type of revision) were generally not correlated. These measures are dissociable in terms of what task environment factors, and pertinent strategic knowledge influenced them. For example, overall quality was related only to syntactic complexity and, surprisingly, total amount or type of revision was not related to quality. The use of the word processor in these studies was critical in permitting realtime analysis of process measures like amount and type of revision. Revision clearly increases when word processing is used as an alternative to handwriting, but the present studies indicate that the revising seldom changes the meaning or holistic quality of the text. In general support of Hayes and Flower’s (1980) theory, writing genre and audience familiarity were shown to influence planning, sentence generation, and reviewing as revealed by writers’ adjustments in syntactic complexity (a product measure) and point of text production revision (a process measure). Writers’ strategic knowledge must include information suggesting that letters to unfamiliar audiences, especially “university presidents,” must be more formal and polished than letters to “Mom and Dad.” This knowledge, then, directs and guides the writer to make more revisions and incorporate syntactically more complex structure when writing to unfamiliar audiences. 524 RansdeiE and Levy It is impo~ant to keep in mind that strategy use based on genre and audience knowledge has been shown to have greater impact on quality than writing tool in several studies (Bridwell, 1980; Collier, 1983; Kellogg & Mueller, 1989; Lutz, 1987). Our evidence suggests that strategic knowledge is dependent on the writer’s ability to coordinate the simultaneous demands of genre, audience, and writing tool familiarity in order to compose successful letters. The clause length analyses revealed that structurally more sophisticated letters were written to the unfamiliar than to the familiar audience. Interestingly, a separate count of the number of task-related ideas created in Experiment 2 (i.e., distinct points made related to the assigned topic) revealed no effect of audience. Thus, these shorter and more sophisticated letters to the unfamili~ audience did not contain fewer task-relevant ideas. They were simply less verbose. The familiar and unfamiliar audiences differed, however, on many dimensions besides familiarity; and in an attempt to maximize a potential effect, these other dimensions were allowed to vary freely. Nevertheless, it should be recognized that while the familiar audience was heterogeneous (including peers and older family members), the unfamiliar audience was not. Even though the experimenter made no references to personal characteristics of the unfamiliar university president, all but one letter was addressed to a man. During debriefings, it was apparent that subjects spontaneously constructed a profile of a university president: a middle-aged man with considerable power, prestige, and literary competence. Having demonstrated a technique capable of documenting that knowledge of one’s audience can influence a composition, we must subsequently determine which combinations of components were most con~buto~ to fully understand the process. For all dependent measures, genre was not as robust an effect as was audience familiarity. Many writers used narrative information in letters that were supposed to focus on argument, though obviously some narrative content is necessary to argue a point. Few, however, introduced argumentation into their narrative letters. When rated genre was evaluated in Experiment 2, we found that subjects tended to create letters containing more narrative elements when writing to a familiar audience than to an unfamiliar one, regardless of their assigned genre. These results may, in part, be due to the difficulty our writers had in producing argumentative text at all. As Ransdell (1989) noted, college students judge w~ting argumentative essays to be a more difficult task than a descriptive narrative. It is not clear yet whether this judged difficulty (and relative paucity of ~gumentation in the present studies) is an inherent difference between these two genre. It is certainly plausible that they are merely indicative of vastly more opportunities to use the narrative vs. the argumentative genre. Certainly everyday conversations tend to be more narrational than expository. And there is a rich literature in cognitive psychology that establishes that cognitive effort needed for successful production of a task varies inversely with knowledge of that task. Although students seemed to take the experimental task seriously, they were writing under time constraints not typically important when they write ordinary business or personal letters. They were also required to write their letters in an atypical public setting. These compro~ses to a study of spontaneous writing in a natural environment were specifically made to enable a ubiquitous adult activity to be examined under controlled conditions. The Hayes and Flower (1980) model has helped to guide writing research for more than a decade, The model has many virtues: it is intuitively plausible, parsi- Process and product 525 moniaus, data based, and sufficiently detailed to generate unidirectional predictions, Xt is an exemplar of Level 4 (process description) inquiry described by Bereiter and ~c~darna~~a (1987). The model must be amended, however, to account for the diss~~ab~e con~ib~tions of writing toof, audience knowledge, rhetorical genre, and writing expertise that we report here to influence both process and products. Whether the Hayes and Flower model can be continuously revised to incorporate new findings and still serve as a heuristic for generating research is an open question. A more important long-term problem for this model is that it lacks We precisely drawn components that would permit it to be proven wrong. Relative to the fields of human memory, attention, and reading, the field of writing is still relatively nascent, As the evolving qualitative and quantitative evidence cm writing processes and products accumulates, some of it appears to be converging. This may signal that the time has come for the emergence of theories of writing having the strength and power of Anderson’s ACT* (1983) or a neural network (e.g., ~c~~e~~and 62 ~urne~h~, 3986) that can hefp guide the next generation of writing research. Acknowledgemenfs - The authors would like to thank Robert Tennyson for serving as editor on this manuscript and two anonymous rcviowers for their helpful comments. NOTES I, Rxty and 44 subjects were orj~~~~Iy run in two separate replication trials, Uriginal experiment served as a factor in initial analyses, and because it did not have any reiiable effects, was collapsed. 2. At the ~ni~~rs~~ of FItida, ~d~r~dua~ must complete six or more courses in which they write at least 6@&3words. T&se were o~m~o~~ly d&med as the courses ~~vo~~~~~“‘extensive” tiring. REFERENCES Anderson, J. R. (1983). The architecture ofcognition. Cambridge, MA: Harvard University Press. Bean, J. C. (1983). Computerized word-processing as an aid to revision. College Composition and Communication, 34, 146-148. Bereiter, C., & Scardamalia, M. (1987). The psychology of w&ten compasition. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc, ~~dw~Il-Bowles, L., Johnson, P., & Brebe, S. (1987). Composing mnd computers: Case studies of experienced writers. In A. Matsn~~s~~ (Ed.), Writing in reaMme: ~~~el~~g production processes. New Jersey: Ablex. Chase, W. G., & Ericsson, K. A. (1982). Skill and working memory. In 0. H, Bower (Ed.), The psychology of learning and memory, Val. 16, l-58, New York: Academic Press. Chi, M. T., Glaser, R., & Rees, E. (1982). Expertise in problem solving, In R. J. Stemberg (Ed.), Advances in the psychology of human intelligence, Vol. 2, Hillsd&, NJ: Lawrence Erlbaum Associates, Inc. C~cbmn-Smith, M. (1991). Word processing and writing in elementary classmoms: A critical review of related literature. Review 0~~~~~~~~~~~ ReseaEh, 63, 107-155. Collier, R. M. (1983). The word processor and revision strategies. Co&ge C~rn~~~~f~o~alsd C~~~~~~~~n~ 34,149-155. 526 Ransdell and Levy Crowhurst, M., & Piche, G. L. (1979). Audience and mode of discourse effects on syntactic complexity in writing at two grade levels. Research in the Teaching ofEnglish, 13(2), 101-109. Daneman, M., Carpenter, P. A., & Just, M. A. (1982). Cognitive progress and reading skills. Advances in Reading Language Research, 1,83-124. Faigley, L., & Witte, S. (1984). Measuring the effects of revisions on text structure. In Research on Written Composition: New directions for teaching (pp. 95-108). Urbana, IL: ERIC Clearinghouse. Hayes, J. R., 8~ Flower, L. S. (1980). Identifying the organization of writing processes. In L. W. Gregg, & E. R. Steinberg (Eds.), Cognitive processes in writing. Hillsdale, NJ: Lawrence Erlbaum Associates. Hayes, J. R., & Flower, L. S. (1986). Writing research and the writer. American Psychologist, 41, 1106-1113. Hillocks, G., Jr. (1986). Research on written composition: New directions for teaching. Urbana, IL: ERIC Clearinghouse. Hunt, K. W. (1983). Sentence combining and the teaching of writing. In The psychology of written language. Chichester: John Wiley and Sons. Joram, E., Woodruff, E., Bryson, M., & Lindsay, P. (1992). The effects of revising with a word processor in written composition. Research in the Teaching of English, 26, 167-193. Kellogg, R. T. (1987). Effects of topic knowledge on the allocation of processing time and cognitive effort to writing processes. Memory and Cognition, l&256-266. Kellogg, R. T., & Mueller, S. (1989). Cognitive tools and thinking pe$ormance: The case of word processors and writing. Paper presented at the annual meeting of the Psychonomics Society, November, 1989. Kirsch, G. (1991). Writing up and down the social ladder: A study of experienced writers composing for contrasting audiences. Research in the Teaching of English, 25, 33-53. Langer, J. A. (1984). The effects of available information on responses to school writing tasks. Research in the Teaching of English, 18, 27-43. Lutz, J. A. (1987). A study of professional and experienced writers revising and editing at the computer and with pen and paper. Research in the Teaching of English, 21(4), 398-421. McClelland, J. L., Rumelhart, D. E., & the PDP Research Group. (1986). Parallel distributed processing: Psychological and biological models, Vol. 2. Cambridge, MA: MIT Press. Nees-Hatlen, V. (1989). Personal communication on procedures for judging holistic quality. Northcraft, G. B., & Neale, M. A. (1987). Experts, amateurs, and real estate: An anchoring and adjustment perspective on property pricing decisions. Organizational Behavior and Human Decision Processes, 39(l), 84-97. Ransdell, S. E. (1989). Producing ideas and text with a word processor. The Computer-Assisted Composition Journal, 4, 22-28. Ransdell, S. E. (1990). Using a real-time replay of students’ word processing to understand and promote better writing. Behavior Research Methods, Instruments, and Computers, 22, 142-144. Redd-Boyd, T. M., & Slater, W. H. (1989). The effects of audience specification on undergraduates’ attitudes, strategies, and writing. Research in the Teaching of English, 23, 77-108. Reed, W. M. (1992). The effects of computer-based writing tasks and mode of discourse on the performance and attitudes of writers of varying abilities. Computers in Human Behavior, 8,97-l 19. Reed, W. M., Burton, J. K., & Kelly, P. P. (1985). The effects of writing ability and mode of discourse on cognitive capacity engagement. Research in the Teaching of English, 19(3), 283-297. Rubin, D. L., & Rafoth, B. A. (1986). Social cognitive abilityas a predictor of the quality of expository and persuasive writing among college freshmen. Research in the Teaching of English, 20(l), 9-21. Sommers, N. (1979). Revision in the composing process: A case study of college freshmen and experienced adult writers. Dissertation Abstracts International, 5374-A. Witte, S. P., & Faigley, L. (1981). Coherence, cohesion, and writing quality. College Composition and Communication, 32, 189-204. APPENDIX A Ho/is tic Quality Rating Dimensions CONTENT OF THE ESSAY Weaknesses - writer uninvolvement with topic and unacknowledged bias; Strengths - engagement with topic and awareness of other views. Process and product 527 PURPOSE/AUDIENCE/TONE: Weaknesses - unclear or unrealized purpose, in appropriate or inconsistent tone; ~~~e~g~~~ - focus and intent clear and consistent and language and tone approp~ate. WORDS/CHOICE AND AR~NGEME~: Weaknesses - awkward or faulty sentences; Strengths - readable, unambiguous sentences. ORGANIZATION AND DEVELOPMENT: Weaknesses - few examples as support, fragmentary thoughts, intent of paragraphing unclear; Strengths - adequate support and elaboration, sense of completeness, and closure and meaningful paragraphing. STYLE: Weaknesses - choppy, difficult to read prose, tendency to play safe with words and ideas; Strengths - fluent, readable prose, occasional willingess to be daring in thought or word. TECHNICAL QUALITY~ECHANICS: Weaknesses - immature sentences, strange idioms, poor gr~rn~, spelling; Strengths - sustained point of view, tenses, gr~matic~ accuracy. Note, Each dimension was rated on a 5-point scale, with 5 being strongest and 1 being weakest.