Coherence in the Narrative and Persuasive Writing of Adolescents
By John Evar Strid
Abstract. This study examined students' narratives and persuasive essays to see if there was a difference in
coherence between the two. This study's sample included Junior High students in the seventh and eighth
grades in a diverse urban public school, whose narrative and argumentative essays were rated for coherence
using Bamberg's (1984) Holistic Coherence Scale. The goal of the study was to look at how genre affects
students' coherence. Thus this study used coherence as a tool to better understand two interesting questions:
how genre affects writing and the importance of genre in instruction. The results also have bearing on the
debate over the relation between cohesion and coherence. The study found that coherence was not
significantly different between persuasive and narrative essays.
Acknowledgements: Rory Donnelly contributed much to the present study and paper.
The origin of coherence as a defining quality of good writing probably dates from the mid 1800s,
when Alexander Bain laid down the first rule of the paragraph: "The bearing of each sentence upon what
precedes shall be explicit and unmistakable." (Quoted in Bamberg, 1984). He also divided prose discourse
into the traditional four discourse modes, and determined that the quality of writing in each mode was a
product of how well the writer handled unity, mass (later known as emphasis), and coherence (Grabe and
Kaplan, 1996 and McCulley, 1985). Bain's analysis was seminal and influenced writing instruction in his
native England as well as the United States well into this century.
Although many parts of his legacy have been supplanted by new concepts and ways of looking at
writing, his division of prose writing into four discourse modes and his notion of coherence seem still to
hold sway in many circles. In fact, both of these 19th century ideas are important to this study, which
examines if coherence varies between genres in seventh and eighth grade students' writing. We will now
examine what work has already been done in this area, looking first at studies relating to coherence and
then at research concerning genre.
Coherence is still thought of as one of the key defining characteristics of the quality of writing and
has been the subject of much recent research. However, since Halliday and Hasan's taxonomy, Cohesion in
English (1976), was published, many have confused their measure of cohesion with coherence. Although
Halliday and Hasan did not explicitly discuss the link between these two terms in their book, they state
their goals in such a way that makes it clear that they view cohesion as a factor in coherence:
If a speaker of English hears or reads a passage of the language which is more
than one sentence in length, he can normally decide without difficulty whether it forms a
unified whole or is just a collection of unrelated sentences. This book is about what makes
the difference between the two.
The word TEXT is used in linguistics to refer to any passage, spoken or written,
of whatever length, that does form a unified whole. We know, as a general rule, whether
any specimen of our language constitutes a TEXT or not...
This suggests that there are objective factors involved -there must be certain
features which are characteristic of texts and not found otherwise... We shall attempt to
identify these, in order to establish what are the properties of texts in English, and what it
is that distinguishes a text from a disconnected sequence of sentences. [p.1]
Thus although the writers did not use the term "coherence," their goals make it clear that they were
grappling with what is generally thought of by that term.
Indeed, many researchers have followed and attempted to sort out how Halliday and Hasan's
taxonomy of cohesion is related to coherence. Generally, most researchers have concluded that coherence
is distinct from and more encompassing than cohesion. Thus cohesion is most commonly viewed as a
subset of coherence -as just one of the factors which create coherence in a text (e.g., Witte and Faigley,
1981; McCulley, 1985; Bamberg, 1984; Fitzgerald and Spiegel, 1986; Morgan and Sellner, 1980).
Therefore, cohesion is usually thought of as the specific surface level ties which create connections
between sentences, while coherence refers to the overall structure, plan or schema that orders the
propositions (e.g. Bamberg, 1984). Therefore, much research has attempted to assess the exact relationship
which exists between these two concepts. While some have been openly critical of any relation between
coherence and Halliday and Hasan's measure of cohesion, others have been more supportive.
McCulley (1985), examining a random sample of the 493 persuasive essay written for the National
Assessment of Educational Progress, found that coherence and cohesion were related, with cohesion
accounting for 53 percent of the variance in coherence. He concluded that "cohesion is a sub-element of
coherence..." [p.278]. Similarly, Fitzgerald and Spiegel (1986 and 1990), in two different studies which
examined the same data (the writing of third and sixth graders) using different methods of analysis, found
coherence and cohesion to be related. In the first study, they utilized Bamberg's (1984 -See below) Holistic
Coherence Scale to establish coherence, while in the second they determined the level of coherence with
Hasan's (1984) Cohesive Harmony Index. Their conclusions in both studies were similar, stating that
cohesion and coherence were related, but that the precise nature of the relationship was uncertain.
This note of uncertainty was also sounded in Witte and Faigley's "Coherence, Cohesion, and
Writing Quality" (1981). Using Halliday and Hasan's taxonomy, they analyzed five student essay rated high
quality using a holistic scale and five rated poor quality. They concluded that there was a relation between
writing quality and cohesion, finding that the well rated essays were denser in cohesive ties. However, they
judged that cohesion is not the same thing as coherence, stating, "Cohesion and coherence interact to a
great degree, but a cohesive text may be only minimally coherent." Thus they viewed cohesion and
coherence as only partially related and they supplied some specific reasons why Halliday and Hasan's
method of analysis does not account for coherence:
They limit cohesion to explicit mechanisms in the text, both the types of cohesive ties that
Halliday and Hasan describe and other elements that bind texts such as parallelism,
consistency of verb tense, and what literary scholars have called 'point of view."
Coherence conditions, on the other hand, allow a text to be understood in a real-world
setting. Halliday and Hasan's theory does not accommodate real-world settings for written
discourse or, consequently, the conditions through which texts become coherent. [p. 199]
In contrast, Tierney and Mosenthal (1983), in a study of twelfth graders' essays ranked holistically
for coherence and then compared for cohesive ties, found no relationship between coherence and cohesion.
They were also critical of Halliday and Hasan, stating that "with the analyses that have been done with
counts of types of cohesive ties, little positive has been stated about the causal relation of Halliday and
Hasan's cohesion concept to textual coherence." [p. 227]
Other writers have been critical of Halliday and Hasan's work on a theoretical level, notably
Morgan and Sellner (1980) and Doyle (1982). Morgan and Sellner critiqued Cohesion in English for not
taking context into account. In Halliday and Hasan's first example ("Wash and core six cooking apples. Put
them into a fireproof dish" [p.2].), they argued that "them" refers back to "six cooking apples" and that the
pronoun makes the text cohesive. Morgan and Sellner argued that them refers to real world objects and not
to elements in a text. Additionally, they argued that the recipe reader creates the coherence using his or her
assumptions or schema for reading a recipe and the meaning that he or she brings to the text. Thus Morgan
and Sellner concluded that Halliday and Hasan mistake the effect of coherence as the cause.
Similarly, Doyle (1982) found fault with Halliday and Hasan's theory for limiting itself to the
surface levels of texts, the consideration of texts after their production and by marking the sentence as a
unit in textual cohesion. She argued studying only the surface level restricted the relationships the
taxonomy can show. Additionally, "Halliday and Hasan limit themselves to a discussion of meaning as it
appears in surface structure; questions of coherence, of the relationships among propositions in the textual
world created by the writer and recreated by the reader, remain unexamined." [p.390]
Bamberg (1984) also criticized the work off Halliday and Hasan, stating that there is a difference
between coherence and cohesion. A text can be cohesive without being coherent as in:
I bought this typewriter in New York. New York is a large city in the USA. Large cities
often have serious financial problems... [p.307]
The lexical ties in this text, consisting of the repetition of terms similar in meaning, do not result in overall
textual coherence, because the piece lacks a semantic scaffold to integrate its disparate points. Drawing on
the work of van Dijk (1980), Bamberg argued that "a text must also have an overall structure, plan, or
schema that orders the propositions." Thus both writers and readers invoke schemas to place and to draw
meaning out of texts in order to make them coherent, something entirely ignored in Cohesion in English.
Therefore, Bamberg set out to develop her own method for measuring coherence, came up with the Holistic
Cohesion Scale, and used it to re-evaluate essays written for the National Assessment of Educational
Progress. Her scale includes cohesive ties as one of seven factors which contribute to coherence (See
Turning to studies which have examined the effect of genre on writing, we find that relatively little
work has been done in this area. Crowhurst has been the most prolific in this area, contributing three
different articles: "Audience and Mode of Discourse Effects on Syntactic Complexity in Writing at Two
Grade Levels" (1979), "Syntactic Complexity in Narration and Argument at Three Grade Levels" (1980),
and "Cohesion in Argument and Narration at Three Grade Levels" (1987). In both the 1979 and 1980
studies, the syntactic complexity of argumentative essays were found to be greater than that of narratives.
In the 1987 study, she found that narratives had a higher use of cohesive ties than did argument, concluding
that "argument is more likely than narration to have generalized statements involving indefinite reference
and thus lower proportionate use of definite reference, that is, of pronouns, demonstratives, and the."
The only other studies of mode are of limited interest. Cox (1986) wrote about the "Cohesion and
Content Organization in the Narrative and Expository Writing of Children," but was more interested in the
differences in writing of good and poor readers than in the differences of writing between the modes that
she studied. Finally, Knudsen (1992) used a variation of Bamberg's (1984) Holistic Coherence Scale
(although she did not give Bamberg any credit) to measure the difference in coherence in student writing
between a complex narrative task and a simple one, finding mixed results. She concluded that there were
two hypothesis to explain the mixed results: "The first hypothesis holds that more complex tasks will
improve the quality of written narrative since students must plan ahead to incorporate all aspects of the task
in the narrative. The second hypothesis suggests that a more complex task will result in poorer written
narratives since the additional demands of the complex text may overload attentional capacity." [p. 7]
Finally, theorists, particularly in Australia, have written about the importance of genre in
instruction (e.g. Grabe and Kaplan, 1996, pp. 131-140). In doing so they are not necessarily arguing for
Bain's division of prose and its appropriateness as a model of writing instruction, but are asserting that
language use is determined by social context and that genre can be a relevant method of varying student's
language output. Thus "Martin, et. al. characterize genres as staged goal-oriented social processes. In
systemic functional linguistics, this amounts to characterizing social context as a system of genres" (Martin,
1993, p.142). Martin then quotes from Bakhtin to show how he presaged this view when writing in 1952:
All the diverse area of human activity involve the use of language. Quite understandably,
the nature of forms of this use are just as diverse as are the areas of human activity...
Language is realized in the form of individual concrete texts (oral and written) by
participants in the various areas of human activity. The texts reflect the specific conditions
and linguistic style, that is the selection of the lexical, phraseological and grammatical
resources of the language, but above all through their compositional structure. All three of
these aspects -thematic content, style, and compositional structure -are inseparably
linked to the whole of the text and are equally determined by the specific nature of
the particular sphere of communication. Each separate text is individual, of course,
but each sphere in which language is used develops its own relatively stable types of these
texts. These we may call speech genres. [The term text has been substituted for utterance
throughout] (As quoted in Martin, 1993, p.142-43)
Therefore, theory holds that differences exist between genres, and one of these differences may be the level
of coherence found in narratives and persuasives written by adolescents.
The Present Study
According to Grabe and Kaplan (1996), "The problem of writing process transfer across tasks and
genres is one which has received relatively little attention..., but which constitutes a serious issue for
writing development." [p.125] We already know that persuasive essays are more syntactically complex than
narratives (Crowhurst and Piche, 1979; Crowhurst, 1980) and that narratives use more cohesive ties than
persuasive essays Crowhurst (1986). But we do not know whether persuasive or narrative essays are more
coherent. Since Crowhurst found that cohesive ties were more prevalent in narratives than in persuasives
using Halliday and Hasan's (1976) taxonomy and cohesion is commonly viewed as a element of coherence,
we predict that narratives have greater coherence than persuasives. Therefore, this study examined
adolescents' writing in order to see if the level of coherence varied across the two modes.
In answering this question, we will thus contribute to the debate about the relationship between
cohesion and coherence. In addition, the study's results will have bearing on the usefulness of genre in
writing instruction. Thus if there is a significant difference in coherence between the two genre types or if
writers' level of coherence is affected by changing genres, we can hypothesize that these genres can be an
important educational tool in certain contexts.
The potential population consisted of fifty-two seventh graders and forty-six eighth graders,
making a total of ninty-seven students. Among the ninty-seven, there was a broad spectrum of abilities and
many ethnic backgrounds. Only thirty-two students handed in both writing samples needed for this study.
Of these thirty-two, twelve or 37.5% were boys and twenty or 62.5% were girls. From the thirty-two, an
additional three were excluded because they handed in a speech for their persuasive essay that they had
been coached on. From the remaining twenty-nine, fifteen students were randomly selected. The resulting
sample had only two boys or 13 percent of the study group.
The students were given a narrative and a persuasive writing assignment two weeks apart. They
were not told that these compositions were any different from similar assignments they had been subject to
since the beginning of the year. The teacher explained what each genre was and showed the students
examples of persuasive and narrative writing. Students were given free selection of topic, although in-class
brainstorming sessions, where the students divided into groups and thought of at least five topics and then
shared them with the class, helped provide them with ideas. Students were asked to bring in a rough draft of
their piece to class the day before it was due and to share it with a friend in peer review. Students were then
asked to rewrite and to turn in their rough draft with their final draft. The student's names were covered
with tape and their compositions were photocopied to maintain anonymity.
Scoring and Method of Analysis
The essays were rated for coherence using Bamberg's Holistic Scale of Coherence. The sample
was given to one judge. The judge was trained in the usage of the scale with a trial run on unrated essays.
The mean and the standard deviation of the scores of all the students was calculated. In addition,
the percentage of essays rated at the different levels were calculated and the percentage of student essays
that were rated the same or different across modes. Finally, a t test was calculated using the mean ratings of
the narratives and the persuasives.
Overall, the persuasive essays were found to be slightly more coherent than the narratives using
the Holistic Coherence Scale.. Means and standard deviations of students' scores for persuasive essays was
2.73, while the mean for narratives was 2.46. The standard deviation for narrative scores was 0.72 and it
was 0.85 for persuasive ratings.
Next, only one or 6.67 percent of the students' narratives was rated
Incomprehensible. While 46.67 percent of the student's narratives were scored Incoherent, 40 percent were
rated Partially Coherent and 6.67 percent were given Fully Coherent. One or 6.67 percent of the persuasive
essay was rated Incomprehensible, one third were given Incoherent, forty-seven percent were scored
Partially Coherent, and twenty percent attained Fully Coherent.
Next, seven out of fifteen students were rated the same on both essays, five were scored higher on
the persuasive than on the narrative, and three were given higher ratings on the narrative than on the
Finally, in the t test of the mean scores of the persuasive and narrative essays, the value of t was
not statistically significant, equaling only 0.88 when it would have had to equal at least 2.048 at 0.05 level
The results of this study show that there is somewhat of a difference between the level of
coherence achieved by adolescents writing narratives and persuasive essays. Overall, more students were
more coherent in writing persuasive essays than they were in narratives. This is the opposite of what we
expected based upon Crowhurst's (1987) finding of more cohesive ties in narratives and the commonly
agreed upon relation between cohesion and coherence (e.g. Witte and Faigley, 1981; McCulley, 1985;
Fitzgerald and Spiegel, 1986 and 1990; Bamberg, 1984). However, the difference in coherence between the
two modes does not meet statistical significance, as established by the t test.
Additionally, the fact that forty-six percent of the students were rated the same in both modes,
thirty-three percent rated better on persuasive than on narrative, and twenty percent rated better on
narrative than on persuasive belies concluding that one mode affected coherence more than the other.
Therefore, we find that neither narratives not persuasives result in more coherent writing for adolescents.
Thus our finding do not correlate with Crowhurst's 1987 study of cohesion and mode and do not support
the contested relation between cohesion and coherence.
However, eight out of the fifteen students did do better writing one genre over the other. Thus
fifty-three percent of the students' writing was affected by genre. This would seem to indicate that genre
does have some effect on coherence, but that the consequence is not consistent for all students. Therefore,
genre would seem to be valuable tool for teaching students how to react to the challenges of writing
different kinds of pieces, as long as the teacher does not expect coherence to be consistently affected for all
In any case, examining the data more closely may reveal some interesting trends. Comparing the
results for individual students in Figure 2, we find that all four of the scores of the students who rated
higher on the narrative than on the persuasive changed by one level: two went from Incoherent to Partially
Coherent, while two went from Partially Coherent to Fully Coherent. In contrast, the scores for students
who were rated higher on the persuasive than on the narrative changed more: two changed from Incoherent
to Partially Coherent, one went from Partially Coherent to Fully Coherent, and two bounded two levels
from Incoherent to Fully Coherent. Beyond the fact that one more student out performed on the persuasive,
this leap goes a long way to explaining the small difference in means.
Looking at the essays of the students who were rated higher in one of the modes will allow us to
understand why their writing was rated more coherent on one writing task than the other. This will permit
us to draw some additional conclusions about coherence and cohesion. The reader may want to consult the
student's compositions in Appendix B.
Starting with the students who were rated higher on the narrative than on the persuasive, we
discover that student number four wrote a very unified and correct description of her room for her narrative
(It was not really a narrative, but it was coherent.), while her persuasive was marked by surface errors and
had no underlying structure. Similarly, student nine had a very coherent narrative recounting the events of
her recent birthday, while her persuasive essay kept referring to "she" without saying who "she" was.
Finally, student five wrote a very unified narrative of what happened when she saw a scary figure in a
window. Her persuasive had some structure, but seemed to flow from one idea to the next without much
Turning to the students who were rated higher in the persuasive than in the narrative, not all of
them encountered similar difficulties. One student, number fourteen, had trouble with her narrative because
she selected a very broad topic, about playing soccer in the park throughout the summer. Therefore, her
narrative had numerous digressions and was very unfocused. In contrast, her persuasive was on one topic
and was much less digressive.
Two students' narratives, numbers one and thirteen, were more focused, having relatively
narrower topics, but they read as lists, with little underlying structure or thought. Therefore, the writers
flowed from detail to detail as they came to mind. Student thirteen's narrative is additionally marred by so
many surface errors that coherence is impeded. Their persuasive essays have some of the same tendencies,
but have less errors and do not read as much like lists, earning them somewhat higher ratings on the
Holistic Coherence Scale.
Finally, two students, number eight and number eleven, wrote extremely well polished persuasive
essays, which had unified topics and no digressions, strong structures that were made known to the reader,
many cohesive ties, practically no surface errors and strong conclusions. In short, their persuasive essays
were fully coherent. However, their narratives did not have the same polish, lacking a clear structure
beyond relating the events in the story. Additionally, both of them lacked conclusions in their narratives,
and one of them had several surface errors.
Therefore, examining all of these students' essays and the reasons some were rated higher than
others, we find that cohesion was never the key factor that made one more coherent than the other. Instead
the presence or lack of mechanical and grammatical errors, underlying structure, reader orientation,
frequent digressions, lack of focus, listing or following associative order, and conclusions were the
ingredients that made essays more or less coherent than the others. The fact that cohesion was not one of
the deciding factors that tipped the scales in determining these essays' coherence goes a long way towards
explaining why the results of this study do not corroborate Crowhurst (1987) and other studies which
correlate cohesion and coherence.
Therefore, this study leaves some issues unresolved which could make for further study. For
example, it might be valuable to replicate this study and add an analysis of cohesive ties between narratives
and persuasives. This would serve the purpose of making the results of this study much stronger. Thus if
Crowhurst's (1987) conclusions were confirmed, meaning narratives have more cohesive ties according to
Halliday and Hasan's taxonomy and there was still not clear agreeing trend for coherence, this would be
even stronger evidence against a firm relation between coherence and cohesion. Thus this result would call
into question the many studies which found a relation between these two concepts (e.g. Witte and Faigley,
1981; McCulley, 1985; Fitzgerald and Spiegel, 1986 and 1990; Bamberg, 1984).
Additionally, it is noteworthy that the only other study which used Bamberg's (1984) Holistic
Coherence Scale to compare two pieces of writing also had mixed results. Thus in both Knudsen's (1992)
study and in the present study, the Holistic Coherence Scale did not yield consistent results, fueling
speculation that there may be problem with this instrument for this kind of research. Because Bamberg
developed the assessment to rapidly rate large numbers of essays in the context of the National Assessment
of Educational Progress, perhaps a more subtle instrument would be more appropriate for the needs of a
study like this one. One possibility could be to break apart the criteria from the holistic scale and to rate the
writing on each attribute, thus converting the holistic scale to an analytic scale.
Finally, one factor which may have a bearing on coherence which was not examined in this study
was the skill level of writers. Thus according to Bereiter and Scardamalia's (1987) model of the writing
process, writers of variant skill levels react in different ways to knowledge telling and knowledge
transforming tasks. Therefore, any future study in this area should also control for this variable.
In conclusion, this study showed that adolescent writers were not consistently more or less
coherent when writing in the persuasive and narrative modes, although sixty percent did show some
change. Thus this study did not find the expected result that narratives would be more coherent than