ArticlePDF Available

Teaching Children to Write: A Meta-analysis of Writing Intervention Research


Abstract and Figures

It has been established that in the Netherlands, as in other countries, a majority of students do not attain the desired level of writing skills at the end of elementary school. Time devoted to writing is limited, and only a minority of schools succeed in effectively teaching writing. An improvement in the way writing is taught in elementary school is clearly required. In order to identify effective instructional practices we conducted a meta-analysis of writing intervention studies aimed at grade 4 to 6 in a regular school setting. Average effect sizes were calculated for ten intervention categories: strategy instruction, text structure instruction, pre-writing activities, peer assistance, grammar instruction, feedback, evaluation, process approach, goal setting, and revision. Five of these categories yielded statistically significant results. Pairwise comparison of these categories revealed that goal setting (ES = 2.03) is the most effective intervention to improve students' writing performance, followed by strategy instruction (ES =.96), text structure instruction (ES =.76), peer assistance (ES =.59), and feedback (ES =.88) respectively. Further research is needed to examine how these interventions can be implemented effectively in classrooms to improve elementary students' writing performance. © Earli. This article is published under Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 Unported license.
Content may be subject to copyright.
Koster, M
A meta-a
JK Utrec
No Deriv
order to
, a
text stru
alysis of writin
onica Koster,
t| The Netherla
: Earli | This arti
tive Works 3.0
, E
an den Be
University |
ity of Amsterd
It has been es
o not attain th
o writin
is li
n improvement
identify effecti
n studies aim
for ten interve
peer assistanc
d revision. Fi
n of these ca
n to improve s
ture instructio
ly. Further rese
in classrooms
: writin
, meta
, De Jon
, P.F.,
intervention r
niversity of Ut
nds – m.p.koste
cle is published
Unported licen
lena Tribu
he Netherlan
am| The Neth
ablished that i
e desired level
ited, and onl
in the way wri
ve instructiona
d at
rade 4 t
tion cate
rammar in
e of these ca
ories reveal
udents’ writin
(ES = .76),
rch is needed
o improve ele
sis, interv
& Van den Ber
search. Journal
echt, Utrecht I
under Creative
n to
f Writi
, Pe
n the Netherla
of writin
a minority of
is tau
ht i
l practices w
6 in a re
: strate
y instru
truction, feedb
ories yielde
d that
oal s
performance, f
peer assistanc
to examine ho
entary students
ntion, com
h, B. (2015) .T
stitute of Lin
Commons Attri
ng Int
ter F. de Jo
ds, as in oth
s at the end o
schools succe
elementary sc
conducted a
school settin
tion, text struct
ack, evaluatio
statistically si
(ES = 2.
llowed by stra
(ES = .59),
these interven
tion, elementar
rch, 7(2), 249-
stics OTS, Tran
r countries, a
elementary sc
d in effectivel
ool is clearly r
e effect
re instruction,
, process appr
nificant result
3) is the mo
y instruction
nd feedback
tions can be i
to write:
10, 3512
ority of
ool. Time
y teachin
equired. In
of writing
sizes were
. Pairwise
t effective
(ES = .96),
ES = .88)
1. Introduction
With the large-scale introduction of computers, tablets, and mobile phones Western
society has rapidly become more literate over the last two decades. E-mail and text
messages are replacing oral face-to-face and telephone communication, increasing
people’s need to be able to communicate adequately in writing. Individuals who do not
sufficiently master the basic skills of writing will eventually encounter serious problems
in participating fully in daily life. More than ever it is essential that children develop
their writing competence at a young age, as writing skills play a crucial role in
educational and occupational success (National Commission on Writing, 2003).
Despite the fact that composition skills are of vital importance for a successful
academic and professional career, it was established that a majority of students in the
Netherlands do not attain the desired level of writing skills (Henkens, 2010). A Dutch
national assessment study demonstrated that at the end of elementary school (grade 6)
most students were not capable of writing texts that sufficiently convey a single, simple
message to a reader (Kühlemeier, Van Til, Feenstra, & Hemker, 2013). Further, this
study showed that students hardly progress in their writing competencies from grade 4
to grade 6. A national writing assessment in the US yielded similar results: of all grade 8
students only one-third performed at or above proficient level (Salahu-Din, Persky, &
Miller, 2008). This is a serious cause for concern, because weaker writers are at a
disadvantage in their secondary school and college years, when writing becomes
increasingly important as a tool for learning (e.g. Bangert-Drowns, Hurley, &
Wilkinson, 2004).
Kühlemeier and colleagues (2013) found that in the Netherlands the time and
attention devoted to writing education is limited at elementary school. At the same time
the Dutch Inspectorate for Education (Henkens, 2010) concluded that only a minority
of schools succeed in effectively teaching writing. Besides, during their own
professional education, Dutch teachers do not receive adequate training in writing
themselves, nor are they sufficiently prepared for teaching writing (Leeuw, 2006; Smits,
2009). Furthermore, it was established that language teaching materials (i.e. textbooks
and teacher manuals) often do not provide sufficient directions for teachers to enable
them to support their students’ writing processes and to give proper feedback
(Stoeldraijer, 2012). It can be concluded that an improvement in the way writing is
taught at elementary school in the Netherlands is clearly required.
Above all, any improvement of the teaching of writing in elementary school must
be based on interventions that have proven to be effective in enhancing the quality of
students’ written texts. The aim of this study was to identify effective instructional
practices for teaching composition to students in the upper grades of elementary
school. An increasing amount of research has been done on writing interventions,
resulting in an accumulation of studies testing various instructional approaches. To gain
insight into which instructional approaches are specifically effective for elementary
students in grade 4 to 6, we conducted a meta-analysis of experimental and quasi-
experimental writing intervention studies aimed at students in the upper elementary
grades. A meta-analysis is the designated method for this purpose, as the magnitude
and the directions of effects of a large number of studies are reviewed in a systematical
In the field of writing research, a number of meta-analyses have already been
conducted. Some of these analyses focused on a specific type of intervention: for
instance, in a review of 39 studies conducted with students from grade 1 to 12, Graham
(2006) found that strategy instruction significantly improved students’ writing
performance. In a meta-analysis on the process approach to writing, Graham and
Sandmel (2011) analyzed 29 studies, involving students grade 1 to 12, and found that
process writing instruction had a significant, but modest, positive effect on the quality
of students’ writing. Furthermore, three meta-analyses (Bangert-Drowns et al., 1993;
Goldberg, Russell, & Cook, 2003; Morphy & Graham, 2012) investigated the effect of
word processing on text quality in grade K to 12, and all found positive effect sizes for
this type of treatment, especially for weaker writers.
So far, there have been three comprehensive meta-analyses of experimental and
quasi-experimental writing intervention studies, investigating multiple treatments: firstly
Hillocks (1984) investigating 60 studies ranging from elementary grades to the first year
of college; secondly Graham and Perin (2007) examining 123 writing intervention
studies with adolescents (grades 4-12); and, thirdly, Graham, McKeown, Kiuhara, and
Harris (2012) analyzed 115 (quasi-) experimental studies involving elementary students,
grade 1 to 6. All three analyses used slightly different intervention categories, due to
differences in the populations under investigation. Despite this, there was substantial
overlap in results. All three meta-analyses consistently found grammar instruction to
have a negative effect on text quality with effect sizes [ES] ranging from -.29 (Hillocks,
1984) to -.41 (Graham et al., 2012). Hillocks (1984) and Graham and Perin (2007) both
found sentence combining (combining simple sentences into more complex ones), with
an ES of .35 and .46 respectively; the study of models (study and imitation of model
pieces of writing), with an ES of .22 and .17; and inquiry (present students with data
and initiate activities designed to help students develop skills or strategies for dealing
with the data in order to write about it), with an ES of .56 and .28, to have a positive
effect on students’ writing performance. Further, Graham and Perin (2007), as well as
Graham and colleagues (2012), found that the process approach to writing, (ES .09 and
.40 respectively); strategy instruction, (ES = 1.03 and 1.02); prewriting activities, (ES of
.42 and .54); peer assistance when writing, (ES = .70 and .89); setting product goals (ES
of 1.00 and .76); and word processing, (ES = .56 and .47), all had a significant positive
impact on text quality. In addition, in their elementary meta-analysis, Graham and
colleagues (2012) identified seven other effective practices to improve the writing of
elementary students writing: feedback (adult and peer), with respective effect sizes of
.80 and .37, the use of creativity and imagery (ES = .70), text structure instruction (ES =
.59), teaching transcription skills (ES = .55), assessing writing (ES = .42), comprehensive
writing programs (ES = .42), and extra writing time (ES = .30).
Rogers and Graham (2008) conducted a meta-analysis of 88 single subject design
studies, and found, consistent with the results of the extensive meta-analyses of
experimental and quasi-experimental studies, that strategy instruction, word processing,
prewriting activities, goal setting, and sentence construction were effective in improving
students’ writing performance. Additionally, Rogers and Graham (2008) found that
reinforcing writing productivity, teaching strategies for editing text, and teaching
strategies for constructing paragraphs were effective for both typical and struggling
writers. In contrast to other findings, Rogers and Graham (2008) found a positive effect
for the teaching of grammar. As a possible explanation for this divergent finding Rogers
and Graham suggested that weaker writers, opposed to typical writers, may have
profited from specific grammar instruction, or that the teaching method (teacher
modeling) may have contributed positively to the effect of grammar instruction.
The meta-analysis that we conducted can be regarded as a refinement of the
previously conducted meta-analyses of writing instruction, as we specifically focused
on effective instructional practices for beginning writers (grade 4-6) in a regular
educational setting. All previous meta-analyses investigating multiple treatments
included a broad range of students: all elementary grades (Graham et al., 2012),
adolescents (Graham & Perin, 2007), or elementary to college students (Hillocks,
1984). We expected, however, that different types of treatment would be effective for
different groups of students. It was our expectation that the effectivity of types of
intervention would differ between elementary students, secondary students, and college
students. Further, we even expected this to differ between lower and upper elementary
students. Bourdin and Fayol (1994) have demonstrated that students until the fourth or
fifth grade in general perform better orally than in writing, when producing narratives.
Their study shows that young students, due to less automation, have to allocate their
cognitive resources mostly to the low-level activities of writing, such as lexical access,
sentence generating, and graphic execution, which interferes with the higher order
skills, such as planning and content generation. Berninger, Yates, Cartwright, Rutberg,
Remy, and Abbott (1992) have shown that in the early elementary grades students’
writing performance is highly dependent on the degree to which the lower level skills
that are conditional to writing are developed. In the upper grades of elementary school
it is expected that these lower level skills have been automatized through maturation
and practice, in such a way that students are able to focus on the composing process
itself (Kress, 1994). It is anticipated that during this stage students will be more sensitive
to instruction and practice in basic composition skills. Therefore we decided, unlike
Graham and colleagues (2012), to exclude studies aimed at the lower levels of
elementary school, and only include studies targeted at students grade 4 to 6.
Further, the prior analyses also included studies targeted at specific groups of
students, for example struggling writers, learning disabled students, bilingual students,
or high-achieving students. In our opinion, one should be cautious to generalize results
from studies targeted at such specific groups to the general population of all students in
a regular school setting, as the instructional needs of these groups are bound to differ.
For instance, bilingual students might need more grammatical and linguistic support,
while the struggling writer might be helped more by instruction in mastering the basics
of writing, whereas gifted students could need more challenging writing tasks and
approaches. For this reason, we chose to include only studies focused at the full range
of students in a regular classroom.
But, above all, none of the previous reviews went beyond summarizing effect sizes
and statistically compared interventions to examine whether they differed significantly
from each other in effectiveness. In that sense they could be considered statistical
reviews more than that they provided answers on the level of differential effectiveness
of specific interventions. With our analysis we expanded the previous meta-analyses by
not only identifying effective interventions, but by also statistically determining their
level of effectiveness by comparison.
Lastly, our meta-analysis can also be considered as an update of the previous body
of meta-analytical research: a quarter of the studies we located were not included in
prior meta-analyses.
In summary, the research question guiding this meta-analysis was: Which
instructional practices effectively improve the writing performance of students in the
upper elementary grades? To answer this question, we systematically reviewed 32
(quasi-) experimental writing intervention studies aimed at students grade 4-6. The
findings from this meta-analysis have important implications for designers of teaching
materials and teacher educators, on how the teaching of composition in upper
elementary education can be improved.
2. Method
2.1 Inclusion criteria and search procedure
In order to be included in the meta-analysis, studies had to meet the following five
criteria. First, the study had to involve students in the upper grades of elementary
school (grade 4-6) in a regular school setting. Studies that were conducted in a special
educational setting or only involving struggling writers were excluded from the analysis.
Second, we only included experimental or quasi-experimental studies in which at least
two instructional conditions were compared: an experimental condition and a control
condition. This could either be a ‘pure’ control condition, in which no extra instruction
was given, or a control condition in which an alternative treatment was provided. As a
consequence, correlational and qualitative studies were excluded from this meta-
analysis. Third, each study had to include a measure of text quality at posttest, as this
provided the best indication of the impact of an intervention on students’ writing
performance. Scores for text quality are based on a reader’s overall impression of the
student’s text, taking into account several factors, such as content, organization,
vocabulary, as well as style and tone. Other outcome measures, such as text length or
students’ motivation were reported only in some of the studies and could therefore not
be included in the meta-analysis. Fourth, to be included in the analysis, studies had to
provide the statistics necessary to compute a weighted effect size. Lastly, only studies
presented in English were included in the meta-analysis.
The studies for this meta-analysis were located by searching the electronic
databases of PsychINFO, ERIC and Google Scholar. For our study, we replicated the
search procedure employed by Graham and colleagues (2012), using the keywords
'writing' or 'composition', combined with keywords indicating the type of
'intervention', such as: assessment, collaborative learning, creativity, dictation, free
writing, genre, goal setting, grammar, handwriting, imagery, inquiry, mechanics,
models, motivation, peer collaboration, peers, planning, pre-writing, process approach,
process writing, self-evaluation, self-monitoring, sentence combining, sentence
construction, spelling, strategies, strategy instruction, summary, technology, word
processing, and word processor. Subsequently, we added the following keywords to
locate potentially promising practices from recent research: editing, feedback,
intervention, modeling, observational learning, outline, outlining revising, and revision.
Further, references of previous meta-analyses, reviews, and obtained papers were
examined for relevant studies.
Databases of theses, dissertations, and conference proceedings were searched for
unpublished studies on the topic. Additionally, a cited reference search of previous
reviews and meta-analyses was conducted in Web of Knowledge to identify relevant
This search procedure yielded approximately 2000 results, of which titles and
abstracts were closely examined. First, we removed all non-intervention studies, as well
as all studies that were not aimed at grades 4-6. Next, we omitted all studies that were
not experimental or quasi-experimental. Subsequently, we removed all studies that
lacked a proper control condition, and finally we excluded all studies only investigating
specific groups of students, such as, for example, struggling writers, learning disabled
students, bilingual students, or high-achieving students. We located 37 studies that met
all inclusion criteria. However, despite this, five studies did not provide the necessary
statistics to calculate effect sizes. We have contacted the authors of these studies to
obtain these statistics, but unfortunately received no reply. Regrettably, these studies
had to be excluded. The procedure described above resulted in the location of 32
studies that were suitable to be included in our meta-analysis.
2.2 Coding procedure
To obtain an adequate description for each study included in the meta-analysis, we
coded the following variables: grade, number of participants, description of
experimental and control condition, publication type (journal/dissertation/report/con-
ference presentation/paper), and the genre of posttest measure (expository/narrative/
informative/persuasive). It should be noted that coding was restricted to posttest
measures, as we used these measures to calculate effect sizes. Further, we coded a
number of variables of which we expected that they could account for heterogeneity in
effect sizes between studies. For this purpose we coded: design of the study (random
assignment or quasi-experimental), attrition (% of total sample), period (in days) and
intensity (in minutes) of intervention, person providing instruction (researcher, teacher,
teaching assistant), and random assignment of teachers to conditions. Due to
considerable differences between the scoring procedures that were used, and
differences in the interpretation of reliability of scoring, it was not possible to administer
one overall reliability score per study. Therefore, we coded aspects of the studies of
which is known that they are related to the reliability of writing quality scores: type of
assessment of writing quality (holistic or analytical), number of writing tasks at posttest
and number of raters assessing the quality of posttest measure (e.g. Rijlaarsdam et al.,
2011). All studies were coded by the first author and a trained assistant, a random
sample of ten studies (one third of the total sample) were coded by both coders, with
97% agreement.
2.3 Categorizing interventions
For the analysis, all studies were thoroughly examined and grouped according to their
focus of intervention. Subsequently, studies with a comparable focus of intervention
were grouped into categories, based on the categories used in previous meta-analyses
(e.g. Graham & Perin, 2007; Graham et al., 2012; Hillocks, 1984). For our study, we
maintained the following categories from these meta-analyses: strategy instruction, text
structure instruction, peer assistance, process approach, feedback, grammar instruction,
and prewriting activities. We decided to use ‘goal setting’ instead of ‘product goals’,
because our sample also included a study involving the setting of process goals, as well
as setting product goals. Two types of intervention in our sample did not fit into the
categories that were used by previous reviews, therefore we added two new categories:
evaluation and revision. This resulted in a total of ten intervention categories, which are
summarized in Table 1. It should be noted that the intervention categories are not
completely mutually exclusive, for instance, prewriting activities and revision also are
components of the process approach and strategy instruction. We classified studies
according to the main focus of instruction as described by the authors. For example,
Bui, Schumaker, and Deshler (2006) characterize their intervention as a strategic
writing program, in which they also apply the process approach to writing. As the
emphasis of this intervention is on teaching students strategies for writing, it was
decided to place this study in the strategy instruction category. Another example of a
study of which the intervention has elements of more than one category is the study of
Wong, Hoskyn, Jai, Ellis, and Watson (2008) which combines self-regulated strategy
development with feedback. As the main intervention under investigation is strategy
instruction, the study was placed in this category, rather than in the category feedback.
Table 1. Description of intervention categories
Strategy instruction involves the explicit teaching of strategies for planning, translating
and revising. The majority of studies in this category uses the Self Regulatory Strategy
Development (SRSD) model of Harris and Graham (1996), in which students are
additionally taught self-regulation strategies to manage the writing process, as well as
declarative and procedural knowledge about writing. Text structure instruction is the
explicit teaching of the structure of a text in a specific genre, such as the organizational
structure of a persuasive essay, the story constituents and interrelations of narrative
texts, or a compare/contrast essay. Peer assistance involves studies where students have
to collaborate during different stages (planning, formulating, or revising) of the writing
process, or where some form of tutoring is applied. Evaluation involves teaching
students how to reflect on and to assess their own work. Most studies in this category
used the 6 (+1) Traits Writing Model, which was developed in the 1980’s in the US
(Northwest Regional Educational Library, 2013). The 6 (+1) Traits Writing Model asks
students to assess their compositions on ideas, organization, voice, word choice,
sentence fluency, conventions, and presentation by using reflective questions and
rubrics. Goal setting involves assigning students goals for their writing before they
begin: either a product goal (e.g. writing paragraphs), or a process goal (e.g. acquiring a
learning strategy). Feedback involves studies in which students receive comments on
(aspects of) their writing, either from the teacher or from a peer. Grammar instruction
involves interventions that are aimed at the construction of correct sentences. Revision
involves studies in which students receive instruction in improving draft versions of
texts. Prewriting activities involve studies that focus on techniques for generating
content and planning, such as brainstorming, or using graphic organizers. The process
approach is a comprehensive intervention where students engage in cycles of planning,
Category Description
Strategy instruction Explicit and systematical teaching of writing strategies
Text structure instruction Explicit teaching of knowledge of the structure of texts
Peer assistance Students engage in joined activities during (parts of) the
writing process
Evaluation Teaching students to evaluate their own work with specified
Goal setting Students are assigned specific product or process goals before
Feedback Students receive comments from others on their writing
Grammar instruction Explicit teaching of grammar and/or construction of sentences
Revision Focus on revising draft versions
Prewriting activities Students engage in activities before writing: generating
Process approach Focus on writing process and subprocesses: planning-writing-
formulating, and revising, and in writing for real audiences with real purposes.
Instruction is often at individual level, tailored to the student’s needs through mini-
lessons, writing conferences, and teachable moments. Further, self-reflection and
evaluation is stressed, to stimulate student’s ownership of their written products.
Students collaborate when writing, in a supportive and nonthreatening writing
environment (Graham & Sandmel, 2011).
There were three studies in our sample, Arter, Spandle, Culham, and Pollard (1993),
Saddler and Graham (2005), and Dejarnette (2008), comparing two intervention
conditions. We calculated an effect size for both interventions and subsequently placed
them in two intervention categories. Finally, a number of studies investigated multiple
conditions, e.g. the study of Schunk and Swartz (1993) investigated the effectiveness of
setting product goals, as well as the effectiveness of setting process goals. In these
instances we calculated separate effect sizes for all conditions.
2.4 Calculation of effect sizes and statistical analysis
For each individual study included in the analysis an effect size was calculated for
writing quality at posttest. If a holistic score was available, then this score was used to
calculate the effect size. If writing quality was scored on separate aspects, such as
organization, ideas, or word choice, separate effect sizes were calculated for each
aspect and subsequently averaged into one single effect size. Means and standard
deviations were used to obtain effect sizes. Effect sizes were calculated using Hedges’ g
(the standardized mean difference) by subtracting the mean performance of the control
group at posttest from the mean performance of the treatment group at posttest, dividing
by pooled standard deviation of the two groups. Hedges’ g provides a slightly better
estimate than Cohens d, especially for smaller sample sizes (Borenstein, Hedges,
Higgins, & Rothstein, 2011).
For the meta-analysis a random effects model was used, as it was assumed that the
true effect varied from study to study, due to differences in participants as well as
differences in interventions and implementation of interventions. Rather than estimating
one true effect size, a random effects model estimates the mean of a distribution of
effects. This allows for generalization to populations beyond the included studies
(Borenstein, et al., 2011). For each treatment category, an average effect size was
calculated as well as the confidence interval and statistical significance of the obtained
effect sizes. In this way the effect of various treatments could be compared.
Additionally, a test of homogeneity was conducted, to determine whether variability in
effect sizes was larger than expected based on sampling error alone. When the
homogeneity test was statistically significant, a moderator analysis was conducted to
determine whether the variability could be explained by identifiable factors, such as
treatment duration, publication type, or grade.
2.5 Description of studies included in the meta-analysis
Table 2 contains a description of all studies included in the analysis and their effect
sizes, grouped per intervention category. The intervention categories are ranked
according to the amount of effect sizes they contain, starting with strategy instruction as
the largest category (11 effect sizes). Within the categories, studies are arranged per
grade, in alphabetical order. For each study the following information is given:
reference, publication type, grade, number of participants, short description of
intervention and control condition, genre of text written at posttest measure, and the
effect size. As can be seen, there are seven categories containing four or less effect
sizes. We acknowledge the fact that these sample sizes do not allow for firm
conclusions. Nevertheless, for the sake of completeness, it was decided to retain these
categories in the analysis, as this would at least provide an indication of the possible
efficacy of these types of interventions. In total, we calculated 55 effect sizes from 32
studies, and divided them into 10 intervention categories.
Table 2. Description of included studies grouped per intervention category
Study Publica-
tion type
Grade N Intervention Genre Effect
Strategy instruction (k=11)
Brunstein & Glaser
J 4 115 Strategy instruction + self-
regulation vs. strategy instruction
N 0.84
Glaser & Brunstein
(2007) 1
J 4 72 Strategy instruction vs. didactic
lessons in composition
N 0.48
Glaser & Brunstein
(2007) 2
J 4 79 Strategy instruction + self-
regulation vs. didactic lessons in
N 1.12
Mason et al. (2012)
J 4 47 Strategy instruction + self-
regulation (TWA + PLANS) vs. no
I 1.13
Bui et al. (2006) J 5 99 Demand Writing Instruction Model
vs. traditional writing instruction
(+Prewriting activities)
n.s. 0.34
Barnes (2013) 1 D 5 178 WISE (Writing In School Every day)
vs. no treatment
N,I,P 0.11
Barnes (2013) 2 D 5 189 WISE + professional development
vs. no treatment
N,I,P 0.33
Mason et al. (2012) J 5 48 Strategy instruction (TWA) vs. no
N 0.81
Study Publica-
tion type
Grade N Intervention Genre Effect
Fidalgo et al. (2015) 6 41 Strategy instruction vs. normal
I 2.11
Torrance et al.
J 6 95 CSRI (Cognitive Self Regulation
Instruction) vs. normal curriculum
I 3.57
Wong et al. (2008) J 6 57 SRSD strategy instruction + CHAIR
+ adult feedback vs. CHAIR +
constant training time
P 0.64
Text structure instruction (k=9)
Fitzgerald & Teasley
J 4 49 Instruction in story constituents
and interrelations vs. dictionary
use and word study
N 1.07
Gordon & Braun
J 5 54 Instruction in narrative structure vs.
instruction in poetry writing
N 0.32
Bean & Steenwyk
(1984) 1
J 6 41 Direct instruction rule-governed vs.
advice to find main ideas
I 1.07
Bean & Steenwyk
(1984) 2
J 6 39 GIST: direct instruction intuitive
approach vs. advice to find main
I 0.84
Crowhurst (1990) J 6 46 Instruction model for persuasion +
writing practice vs. group
discussion activities
I 1.11
Crowhurst (1991) 1 J 6 50 Instruction model for persuasion +
writing practice vs. reading novels
and writing book reports
P 1.10
Crowhurst (1991) 2 J 6 50 Instruction model for persuasion +
reading practices vs. reading
novels and writing book reports
P 0.78
Crowhurst (1991) 3 J 6 50 One lesson persuasion vs. reading
novels and writing book reports
P 0.34
Raphael & Kirschner
C 6 45 Instruction compare-contrast text
structure vs. normal curriculum
I 0.26
Peer assistance (k=9)
Paquette (2008) J 4 50 6 + 1 Traits model with cross-age
tutoring vs. no extra instruction
(+ Evaluation)
n.s. 1.27
Study Publica-
tion type
Grade N Intervention Genre Effect
Puma et al. (2007) 1 R 4 124
Writing Wings (cooperative
writing) vs. normal curriculum
N,I 0.07
Saddler & Graham
(2005) 1
J 4 44 Sentence combining with peer
assistance vs. grammar instruction
N 1.66
Puma et al. (2007) 2 R 5 347 Writing Wings (cooperative
writing) vs. normal curriculum
N,I 0.03
Yarrow & Topping
(2001) 1
J 5 14 Metacognitive strategy instruction
with peer assistance (tutor) vs.
metacognitive strategy instruction
with no interaction
N 0.70
Yarrow & Topping
(2001) 2
J 5 12 Metacognitive strategy instruction
with peer assistance (tutee) vs.
metacognitive strategy instruction
with no interaction
N 0.52
Brakel Olson (1990)
J 6 41 Writing lessons + peer partner vs.
writing lessons only
N 0.42
Hoogeveen (2013) 1 D 6 96 Specific genre knowledge + peer
response vs. no extra instruction
N,E 1.11
Hoogeveen (2013) 2 D 6 93 General aspects of communicative
writing + peer response vs. no
extra instruction
N,E 0.30
Evaluation (k=7)
Collopy (2009) J 4 100 6 Traits writing model vs. no extra
N 0.31
Paquette (2008) J 4 50 6 + 1 Traits model with cross-age
tutoring vs. no extra instruction
(+ Peer assistance)
n.s. 1.27
Tienken & Achilles
J 4 98 Skills and strategies to self-assess
writing vs. no extra instruction
N 0.41
Ross et al. (1999) J 4/5/6 296 Self-evaluation with rubrics +
teacher feedback vs. normal
curriculum development
N 0.74
Arter et al. (1994)1 C 5 132 6 Traits writing model vs.
observation (normal curriculum)
(+ Process approach)
E,N 0.20
Study Publica-
tion type
Grade N Intervention Genre Effect
DeJarnette (2008) D 5 131 6 + 1 Traits writing model vs.
Writing workshop
N 0.73
Coe et al. (2011) R 5 413
6 Traits writing model vs. no extra
E 0.01
Goal setting (k=6)
Schunk & Swartz
(1993) 2
J 4 20 Process goal + progress feedback
vs. general goal (+ Feedback)
E,N,I 3.03
Schunk & Swartz
(1993) 2
J 4 20 Process goal vs. general goal E,N,I 2,62
Schunk & Swartz
(1993) 2
J 4 20 Product goal vs. general goal E,N,I 1,05
Schunk & Swartz
(1993) 1
J 5 30 Process goal + progress feedback
vs. general goal (+ Feedback)
E,N,I 3.15
Schunk & Swartz
(1993) 1
J 5 30 Process goal vs. general goal E,N,I 2.66
Schunk & Swartz
(1993) 1
J 5 30 Product goal vs. general goal E,N,I 1.65
Feedback (k=4)
Schunk & Swartz
(1993) 2
J 4 20 Process goal + progress feedback
vs. general goal (+ Goal setting)
E,N,I 3.03
Schunk & Swartz
(1993) 1
J 5 30 Process goal + progress feedback
vs. general goal (+ Goal setting)
E,N,I 3.15
Holliway (2004) 1 J 5 55 Feedback + rating vs. one sentence E 0.84
Holliway (2004) 1 J 5 48 Feedback + reading as the reader
vs. one sentence feedback
E 0.69
Grammar instruction (k=4)
Saddler & Graham
(2005) 1
J 4 44 Grammar instruction vs. sentence
combining with peer assistance
N -1.66
Gein (1991) 1 D 4 109 School grammar vs. direct writing E,N -0.05
Gein (1991) 2 D 4 110 Sentence construction vs. direct
E,N 0.06
Gein (1991) 3 D 4 111 School grammar vs. sentence
E,N -0.11
Study Publica-
tion type
Grade N Intervention Genre Effect
Revision (k=3)
Brakel Olson (1990)
J 6 40 Revision instruction vs. no extra
N 0.04
Brakel Olson (1990)
J 6 37 Revision instruction + peer partner
vs. no extra instruction (+ Peer
N 0.85
Fitzgerald &
Markham (1987)
J 6 30 Revision instruction vs. reading
good literature
N 0.89
Prewriting activities (k=3)
Brodney et al. (1999)
J 5 51 Reading combined with prewriting
vs. no extra instruction
E 0.93
Brodney et al. (1999)
J 5 49 Prewriting only vs. no extra
E 0.17
Bui et al. (2006) J 5 99 Demand Writing Instruction Model
vs. traditional instruction
(+ Strategy instruction)
n.s. 0.34
Process approach (k=3)
Arter et al. (1994) C
5 132 Process approach vs. 6 Traits
model (+ Evaluation)
E,N -0.20
DeJarnette (2008) 2 D 5 131 Writing workshop vs. 6 + 1 Traits
writing model (+Evaluation)
N -0.73
Varble (1990) J 6 128 Whole language group vs.
traditional language instruction
I 0.16
Note: For study, numbers behind the references indicate that effect sizes were calculated for
multiple conditions, or groups; these effect sizes are reported separately. For Publication type, J:
Journal, D: Dissertation, R: Report, C: Conference presentation, P: Paper. For Genre, N: Narrative,
E: Expository, I: Informative, P: Persuasive, n.s.: not specified. When a study is included in another
category as well, this is mentioned in parentheses.
3. Results
First, a random effects model was used to obtain an overall average effect size for all
studies included in the meta-analysis. This overall effect size was g = .72, with a 95%
confidence interval of [.49 - .94]. As effect sizes are highly dependent on study
characteristics, additional analysis was needed to establish whether the various effect
sizes together in the sample provide a proper estimate of the effect size in the
population. This can be determined by conducting a homogeneity test. This test
indicates if the variability in effect sizes is larger than the expected variability based on
sampling error alone. As the studies in our sample varied widely in focus and approach,
we expected significant heterogeneity, which was confirmed by the homogeneity test:
Q = 511.51, df = 54, p < .001. This indicated that a common effect size for the total
sample of studies could not be assumed.
First, we investigated possible publication bias by conducting a moderator analysis
with publication type as moderator on all studies in the meta-analysis. This analysis
yielded no significant result (p = .22), indicating that the effect sizes of studies
published in peer reviewed journals did not differ systematically in their effect sizes
from studies from other publication types. The next step in our analysis was to examine
the effectiveness of the various intervention categories, by including these 10 categories
in our model as explanatory variables. The inclusion of the intervention categories
significantly improved the model, according to a likelihood ratio test, with Χ2 = 19.69,
df = 9, p < .001. This means that differences in effect sizes were (at least partly)
explained by the type of intervention.
Table 3. Summary of statistics for intervention categories
Intervention N Average SE 95% Confidence Heterogeneity
Lower Upper Q I2
Strategy instruction 11 0.96 *** 0.19 0.59 1.33 109.99*** 94.30
Text structure 9 0.76 *** 0.21 0.34 1.18 11.91 33.87
Peer assistance 9 0.59 ** 0.21 0.17 1.01 56.05*** 89.83
Evaluation 7 0.43 0.23 -0.01 0.87 66.56*** 87.57
Goal setting 6 2.03 *** 0.33 1.37 2.68 13.47* 62.61
Feedback 4 0.88 * 0.38 0.14 1.61 25.08*** 91.08
Grammar 4 -0.37 0.30 -0.97 0.22 20.16*** 91.84
Revision 3 0.58 0.38 -0.17 1.33 4.14 51.59
Prewriting activities 3 0.13 0.36 -0.58 0.85 3.91 48.57
Process approach 3 -0.25 0.34 -0.92 0.41 12.78** 84.58
Note: *** p < .001, ** p < .01, * p < .05
Table 3 gives the summary statistics for all intervention categories, presented in the
same order as Table 2. These statistics include, per intervention category, the number of
effect sizes, the average effect size and standard error, the 95% confidence interval, and
the heterogeneity statistics Q (test statistic for heterogeneity) and I2 (percentage of total
As can be seen in Table 3, we found two negative effects, for grammar instruction and
the process approach. These interventions did not improve the quality of students’
writing. However, all other main effects were positive. Of these positive effects, five
main effects significantly deviated from zero. These were, in order of effect size: goal
setting, strategy instruction, feedback, text structure instruction, and peer assistance.
Post-hoc analysis was conducted by a contrast analysis in which all interventions were
compared pairwise. Results from these analyses showed that goal setting was by far the
most effective intervention (Χ2 36.81, df = 1, p < .001). However, as can be seen in
Table 2, all effect sizes in the category goal setting were calculated from one study in
which multiple conditions and grades were compared (Schunk & Swartz, 1993). This
result should therefore be interpreted with caution. Goal setting was followed by
strategy instruction (Χ2 26.06, df = 1, p < .001), text structure instruction (Χ2 12.82,
df = 1, p < .001), and peer assistance (Χ2 7.64, df = 1, p = .006) respectively. These
three categories were all based on nine or more effect sizes from different studies.
Feedback also proved to be an effective intervention, however, not more effective than
prewriting activities.
The homogeneity test indicated that there was still a significant amount of residual
heterogeneity in the sample (QE = 283.18, df = 45, p < .001). Therefore, we inspected
the funnel plot (see Figure 1) to locate outliers that could be a potential source of
heterogeneity. A funnel plot is a scatterplot of the intervention effect against a measure
of study size. In the funnel plot in Figure 1 the residuals of the model with the
intervention categories as explanatory variables were plotted against the standard error.
The straight lines in Figure 1 define the region in which 95% of the studies was
expected, in the absence of homogeneity. It can be seen that the studies were more or
less symmetrically spread around the overall average effect size, and that most points
were located in the region between the straight lines. This was an indication that there
was no systematic heterogeneity in our sample. Two outliers (6.25% of the total
sample) were located. The forest plot that we subsequently created (see Appendix),
identified these outliers to be the studies of Torrance et al. (2007), and Saddler and
Graham (2005). The effect size in the study of Torrance et al. (2007) was
underestimated in the analysis whereas the observed effect size in the study of Saddler
and Graham (2005) was smaller than expected, which meant that the effect size of the
first study was larger, whereas in the latter study the effect size was smaller than in
comparable studies (see also Figure 1). The analysis was repeated without these studies,
but the outcome of this analysis did not significantly differ from the previous analysis
(Χ2 = 3.61, df = 2, p = 0.16). Hence, it was decided to maintain these studies, and to
retain the previously estimated model for further analysis.
Figure 1: Funnel plot of final model.
Subsequently, a moderator analysis was conducted to examine whether the variability
between studies could be attributed to one or more identifiable factors. We examined
whether there were systematical differences in effect sizes between studies with a
proper control condition and studies comparing different intervention conditions. In six
intervention categories there were one or more studies without a no extra instruction
control condition. Contrary to expectations, the inclusion of control condition as a
moderating variable did not result in a significant reduction of residual heterogeneity
(QE = 220.37, df = 37, p < .001), and for none of the intervention categories the
parameter estimate for control condition was significant (p-values ranging from .29 to
.90). Next, grade, duration of intervention, type of assessment of writing quality (holistic
or analytical), number of writing tasks in posttest, and number of raters assessing the
quality of posttest measure were considered as moderating factors. None of these
factors significantly reduced the heterogeneity between studies in the total sample.
Residual Value
Standard Error
0.784 0.588 0.392 0.196 0.000
-2.00 -1.00 0.00 1.00 2.00 3.00
In the next step of the analysis, closer examination of the intervention categories
separately revealed no significant heterogeneity in four categories: text structure
instruction, process approach, revision, and prewriting (p-values ranging from .08 to
.16). We further investigated the heterogeneity within the remaining intervention
categories. This analysis was limited to categories containing more than five effect
sizes, i.e. strategy instruction, peer assistance, evaluation and goal setting, as in the
smaller categories the heterogeneity can largely be attributed to differences between
individual studies. In the larger categories systematic factors may have caused
heterogeneity, and this was examined by performing a moderator analysis on the
separate categories with grade, duration of intervention, type of assessment of writing
quality (holistic or analytical), number of writing tasks in posttest, and number of raters
assessing the quality of posttest measure as potential moderators.
In strategy instruction, grade appeared to be a significant moderator: effect sizes
were systematically larger in grade 6 (2.19) than in either grade 4 or 5 (0.59). Further,
we found that effect sizes in this category were smaller (-0.86) for studies in which text
quality at posttest was assessed analytically compared to studies in which holistic
assessment was applied. In the category evaluation, genre of posttest was a significant
moderator: effect sizes were smaller (-0.11) for expository texts. In the category peer
assistance, heterogeneity could largely be attributed to one large study (Puma et al.,
2007) with a relatively low effect size. In goal setting, heterogeneity could be attributed
to differences between conditions.
However, from the 95% confidence interval statistics reported in Table 3 can be
concluded that, despite significant heterogeneity within the categories of interventions
that significantly improve writing proficiency, the effects in these categories were still
largely positive, even at the lower bound of the confidence interval.
4. Discussion
4.1 Effective interventions to improve elementary students’ writing
It has been established that, in the Netherlands, the way writing is taught in elementary
school needs to be improved. The aim of this meta-analysis was to identify evidence-
based effective instructional practices for teaching writing to students in grade 4 to 6.
To determine this, we calculated average effect sizes for 10 types of interventions. The
results show that the most effective interventions to improve students’ writing are, in
order of effect sizes: goal setting, strategy instruction, text structure instruction,
feedback, and peer assistance. Post hoc analysis demonstrates that goal setting is the
most effective intervention, followed by strategy instruction, text structure instruction,
peer assistance, and feedback. This is in line with the findings of recent previous
reviews (Graham & Perin, 2007; Graham et al., 2012), even though we limited our
analysis to students in grade 4 to 6 in a regular educational setting. However, our
findings are corroborated by statistical analysis.
The results of our analysis show that goal setting was by far the most effective
intervention. However, as stated before, it should be noted that all effect sizes in this
category come from one (twenty year old) study (Schunk & Swartz, 1993), comparing
multiple conditions and multiple grades. Thus, these results only allow for tentative
conclusions. Support for the positive effect of setting product goals can be found in
previous meta-analyses (Graham & Perin, 2007; Graham et al., 2012) albeit for (partly)
different populations of students (special needs learners, struggling writers, and slightly
older students). This indicates that setting goals could help to improve students’ writing.
Strategy instruction is the next effective intervention. Strategy instruction is the
largest intervention category in our analysis, which allows for robust conclusions. Of all
types of intervention, strategy instruction is by far the most investigated. It should be
noted that that the majority of studies in this category examined the self-regulated
strategy development (SRSD) approach of Harris and Graham (1996) to strategy
instruction or a variation thereof. The SRSD approach seems to have developed into the
‘standard’ in strategy instruction, which is hardly surprising as studies examining SRSD
invariably yield large effect sizes. Previous meta-analyses (Graham, 2006; Graham &
Perin, 2007; Graham et al., 2012) also found SRSD to be a highly effective intervention
for all types of learners (struggling writers, learning disabled, average, gifted) in a wide
range of grades (grade 2 to 10). A subsequent moderator analysis, which we performed
in all categories containing more than five effect sizes from different studies, shows that,
in our sample, in grade 6 the (average) effect of strategy instruction appears to be much
higher than in either grade 4 or 5. A possible explanation for this finding may be that in
grade 6 students’ lower level skills have been developed to such an extent that they
profit the most from the explicit teaching of writing strategies. Further, we find that
effect sizes in this category are smaller in studies where text quality is assessed
analytically, compared to studies in which holistic assessment is used. In analytical
assessment scoring rubrics are used: a set of criteria and standards that are linked to the
learning objectives of the task at hand. Therefore, analytical assessments are more task-
specific than a holistic assessment, which makes them harder to generalize to writing
proficiency (Schoonen, 2005; Rijlaarsdam et al., 2011). As all different aspects of a text
are evaluated separately, and subsequently combined into one final total score,
analytical scores tend to be lower than holistic scores (Schoonen, 2005).
The next effective intervention category is text structure instruction. This category is
a homogeneous sample of studies. The studies in this category investigate the effect of
explicit teaching of (elements of) text structure, in different types of texts: narrative,
persuasive, and compare-contrast texts. In all studies in this category the explicit
teaching of text structure leads to a significant improvement of students’ writing
Text structure instruction is followed by peer assistance. Peer assistance is a diverse
category: collaboration between students is applied in different phases of the writing
process, with diverse types of interventions. As can be seen in Table 2, the effect of
peer assistance depends on how it is applied, and on the focus of the intervention.
Studies with mainly cooperative writing (e.g. Puma et al., 2007) have smaller effects
than studies combining peer assistance with more targeted types of interventions, such
as the teaching of specific genre knowledge (Hoogeveen, 2013) or sentence combining
(Saddler & Graham, 2005). Peer tutoring is also an effective practice to improve
students’ writing, as is shown by the study of Yarrow and Topping (2001). This study
further shows that the writing scores of tutors improved more than those of the tutees.
An explanation for this result may be that students learn more from explaining the
material to others: you can only adequately explain something if you understand it
With only four effect sizes from two studies, feedback is one of the smaller
intervention categories. Although seemingly effective, more research is needed to allow
for more robust conclusions, as feedback can take many forms (e.g. peer feedback vs.
teacher feedback) and can be applied in different ways (e.g. product-focused vs.
process-focused). Further research should examine how and in what form feedback can
be applied in teaching writing to improve students’ writing performance.
Grammar instruction and the process approach to writing yield negative average
effect sizes. The negative effect for grammar instruction confirms the findings in
previous meta-analyses (Graham & Perin, 2007; Graham et al., 2012; Hillocks, 1984).
Apparently, attention for the construction of correct sentences does not lead to
improvement in text quality. This may be due to lack of transfer effects: when grammar
is taught in isolation, and not in a ‘real’ writing context, it may not be clear to students
how to apply what they learned when writing a text.
The negative effect for process approach may be explained by several factors. First,
it is a small, but nevertheless homogeneous, intervention category of only three studies.
In two out of these three studies, process approach is the control condition, thus
compared with another (in this case: more effective) intervention type. We suspected
that this could have resulted in lower effect sizes than when process approach would
have been compared to a ‘pure’ control group. However, our suspicions were not
confirmed by subsequent analysis with type of control condition as a moderator. There
are several possible explanations for this result: the most straightforward one is that
there are indeed no differences, but it can also be that our sample is too small and
therefore lacking power to reveal systematic differences. However, it can also be that
the process approach is too comprehensive for beginning writers: working on too many
aspects at the same time. Beginning writers may profit more from a targeted
intervention, such as text structure or strategy instruction. It must be noted that Graham
and Perin (2007) found a (small) positive effect, for the process approach in their meta-
analysis for adolescent students. This might indicate that the process approach is an
effective approach for teaching writing to more experienced writers, but that this
approach is less suitable for beginning writers.
4.2 Limitations of the study
We recognize the fact that some categories were small ( 4 effect sizes) and therefore
only allow for tentative conclusions on the overall effectivity of these intervention
categories. Nevertheless, since we wanted to obtain as much information as possible
from the available data, these categories were included in the analysis, to examine their
potential effectiveness.
A complicating factor in interpreting the results of the analysis is the fact that there
was a considerable heterogeneity between studies that could not fully be accounted for
by identifiable factors. However, it should be noted that the heterogeneity is
overestimated due to the amount of small studies in our sample. A large amount of
small studies in a category results in considerably more heterogeneity between studies,
whereas in larger studies there is more heterogeneity within the study, and less between
studies. In our sample, we see that the heterogeneity in the smaller categories is often
caused by differences between individual studies. For instance, variation between
studies may have been caused by differences in operationalization, such as the
materials that were used, and the instruction that was given. Further, the number and
nature of assignments that students had to work on varied considerably: from one
writing task in one genre, to several writing tasks in various genres. The period of
intervention varied even more: from one day to one year. A complicating factor in the
analysis is that key aspects, such as the control condition, the nature of the posttest, the
exact period of intervention, and the exact time spent on the intervention were not
always described explicitly, which made it troublesome to code for these variables.
These aspects can contribute to heterogeneity, but they cannot be included in a meta-
analysis in a meaningful way if they are not reported accurately.
4.3 Suggestions for further research
From our study it becomes clear that there is not much writing intervention research
conducted for students in the upper grades of elementary school. We can conclude that
more research in this area is clearly needed. Some of the intervention categories in our
meta-analysis were too small to draw firm conclusions about their effectiveness.
Especially in these categories more research should be conducted. Particularly the
effectiveness of goal setting needs further investigation, as our results indicate that it
might be very effective in improving writing. It would certainly be worthwhile to
investigate whether the positive results of the study of Schunk and Swartz (1993) can be
replicated in other studies. But also feedback and prewriting activities need to be
studied more closely. Additionally it should be examined if a combination of highly
effective interventions will lead to even better student outcomes. In other words: is it
worthwhile to add one highly effective intervention to another highly effective
intervention, or does this only lead to marginal improvement? Further, other types of
interventions, and new approaches should be developed and tested.
Furthermore, of the studies in our sample, 34% employs a posttest-only design and
47% a pretest-posttest design in which the effect is measured directly at the end of an
intervention. However, to make substantiated claims about the effectiveness of an
intervention, a delayed posttest should be included to measure retention. Often, the
posttest closely resembles what is taught during the intervention, which may lead to an
overestimation of the effects. A delayed posttest could provide more information on the
long-term effects of interventions on students’ writing. Therefore, to make any claims
about the ‘real’ effectiveness of interventions, delayed posttest data are essential.
Unfortunately, this is still not common practice in intervention research.
4.4 Recommendations for teaching
This meta-analysis provides some valuable clues as to what works in teaching writing.
Certainly more in-depth research is needed into what specifically works and what not,
but we were already able to identify promising interventions for successfully teaching
writing to students in the upper grades of elementary education. On the basis of our
results, we must conclude that, to successfully improve the quality of writing of
beginning writers, the writing curriculum should include goal setting, strategy
instruction, text structure instruction, feedback and peer interaction. Setting process
goals, such as learning to apply a certain strategy, was highly effective. Strategy
instruction was more effective when combined with teaching self-regulatory skills.
Overall, we found that specific, targeted interventions, such as explicit instruction in
applying strategies or how to structure a text were particularly effective for elementary
students. What we still do not know, is what the ideal instructional program for
teaching composition skills should look like: which materials should we use, how
much students have to write, how much practice students need, how we support the
students’ writing process, how we give appropriate feedback, and so on. In that respect,
this analysis provides only rough guidelines for teaching, not a ready to use panacea.
To determine what really works, extensive testing in classrooms is still needed.
This research has been supported by the Netherlands Organization for Scientific
Research (NWO), grant 411-11-859.
As this information was available for all studies, we restricted ourselves to calculating effect sizes
for posttest only. Pretest information was available for 66% of the studies in our sample, whereas
19% of the studies also included a delayed posttest measurement. Pretest measures were only
coded to verify whether there were pre-intervention differences between conditions, as most
studies had a quasi-experimental design.
References marked with * are studies included in the meta-analysis
*Arter, J. A., Spandel, V., Culham, R., & Pollard, J. (1994). The Impact of Training Students To Be
Self-Assessors of Writing. New Orleans. Paper presented at AERA.
Bangert-Drowns, R.L., Hurley, M.M.,& Wilkinson, B. (2004). The effects of school-based Writing-
to-Learn interventions on academic achievement: A meta-analysis. Review of Educational
Research, 74, 29-58.
*Barnes, J. C. (2013). The Effects of a Writing Intervention on Fifth-Grade Student
Achievement (Doctoral dissertation).
*Bean, T. W., & Steenwyk, F. L. (1984). The effect of three forms of summarization instruction on
sixth graders' summary writing and comprehension. Journal of Literacy Research, 16(4), 297-
Berninger, V., Yates, C. Cartwright, A., Rutberg, J., Remy, E., & Abbott, R. (1992). Lower-level
developmental skills in beginning writing. Reading and Writing: An Interdisciplinary Journal, 4,
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2011). Introduction to meta-
analysis. West Sussex: Wiley.
Bourdin, B., & Fayol, M. (1994). Is written language production more difficult than oral language
production? A working memory approach. International Journal of Psychology, 29(5), 591-
*Brakel Olson, V.L. (1990). The revising processes of sixth-grade writers with and without peer
feedback. The Journal of Educational Research, 84(1), 22-29.
*Brodney, B., Reeves, C., & Kazelskis, R. (1999). Selected prewriting treatments: Effects on
expository compositions written by fifth-grade students. The Journal of Experimental
Education, 68(1), 5-20.
*Brunstein, J. C., & Glaser, C. (2011). Testing a path-analytic mediation model of how self-
regulated writing strategies improve fourth graders' composition skills: A randomized
controlled trial. Journal of Educational Psychology, 103(4), 922-938.
*Bui, Y. N., Schumaker, J. B., & Deshler, D. D. (2006). The Effects of a Strategic Writing Program
for Students with and without Learning Disabilities in Inclusive FifthGrade Classes. Learning
Disabilities Research & Practice, 21(4), 244-260.
*Coe, M., Hanita, M., Nishioka, V., & Smiley, R. (2011). An Investigation of the Impact of the 6+ 1
Trait Writing Model on Grade 5 Student Writing Achievement. Final Report. NCEE 2012-
4010. National Center for Education Evaluation and Regional Assistance.
*Collopy, R. M. (2008). Professional development and student growth in writing. Journal of
Research in Childhood Education, 23(2), 163-178.
*Crowhurst, M. (1990). Reading/writing relationships: An intervention study. Canadian Journal of
Education/Revue canadienne de l'éducation, 15(2), 155-172.
*Crowhurst, M. (1991). Interrelationships between reading and writing persuasive
discourse. Research in the Teaching of English, 25(3), 314-338.
*DeJarnette, N. K. (2008). Effect of the 6+ 1 Trait Writing Model on Student Writing Achievement.
*Fidalgo, R., Torrance, M., Rijlaarsdam, G., Bergh, H. van den, Álvarez, M.L. (2015). Strategy-
focused writing instruction: just observing and reflecting on a model benefits 6th grade
students. Contemporary Educational Psychology, 41, 37-50.
*Fitzgerald, J., & Teasley, A. B. (1986). Effects of instruction in narrative structure on children's
writing. Journal of educational psychology, 78(6), 424-432.
*Fitzgerald, J., & Markham, L. R. (1987). Teaching children about revision in writing. Cognition
and Instruction, 4(1), 3-24.
*Gein, J. van de (1991). The sense of sentences. A study into the effects of grammar instruction
upon junior writing. Doctoral dissertation. Utrecht: Utrecht University.
*Glaser, C., & Brunstein, J. C. (2007). Improving fourth-grade students' composition skills: Effects
of strategy instruction and self-regulation procedures. Journal of Educational Psychology, 99(2),
Goldberg, A., Russell, M., & Cook, A. (2003). The effect of computers on student writing: A meta-
analysis of studies from 1992 to 2002. Journal of Technology, Learning, and Assessment, 2(1).
*Gordon, C. J., & Braun, C. (1986). Mental processes in reading and writing: A critical look at self-
reports as supportive data. The Journal of Educational Research, 79(5), 292-301.
Graham, S. (2006). Strategy instruction and the teaching of writing. In C. A. MacArthur, S. Graham
& J. Fitzgerald (Eds.), Handbook of writing research (1st ed., pp. 187-207). New York: The
Guilford Press.
Graham, S., & Perin, D. (2007). A meta-analysis of writing instruction for adolescent students.
Journal of Educational Psychology, 99(3), 445-476.
Graham, S., & Sandmel, K. (2011). The process writing approach: A meta-analysis. The Journal of
Educational Research, 104(6), 396-407.
Graham, S., McKeown, D., Kiuhara, S., & Harris, K. R. (2012). A meta-analysis of writing
instruction for students in the elementary grades. Journal of Educational Psychology, 104(4),
Harris, K., & Graham, S. (1996). Making the writing process work: Strategies for composition and
self-regulation (2nd ed.). Cambridge, MA: Brookline Books.
Henkens, L. (2010). Het onderwijs in het schrijven van teksten [Education in text writing] (pp. 1–
58). Utrecht: Inspectie van het Onderwijs.
Hillocks, G. (1984). What works in teaching composition: A meta-analysis of experimental
treatment studies. American Journal of Education, 93(1), 133-170.
*Holliway, D. R. (2004). Through the eyes of my reader: A strategy for improving audience
perspective in children's descriptive writing. Journal of Research in Childhood
Education, 18(4), 334-349.
*Hoogeveen, M. (2013). Writing with Peer Response using Genre Knowledge. Doctoral
Dissertation. SLO: Enschede.
*Knudson, R. E. (1991). Effects of instructional strategies, grade, and sex on students' persuasive
writing. The Journal of Experimental Educational, 141-152.
Kress, G. (1994). Learning to write. London: Routledge.
Kühlemeier, H., Van Til, A., Feenstra, H., & Hemker, B.(2013). Balans van de schrijfvaardigheid in
het basis- en speciaal basisonderwijs 2. Periodieke Peiling van het Onderwijsniveau (No. 53)
[Present state of writing competency in elementary and special education 2. Periodical
assessment of the level of education]. Arnhem: Cito.
Leeuw, B. van der (2006). Schrijftaken in de lerarenopleiding. Een etnografie van
onderwijsvernieuwing. [Written assignments in teacher training: An ethnography of
educational reform]. Doctoral dissertation. Utrecht: Utrecht University.
*Mason, L. H., Davison, M. D., Hammer, C. S., Miller, C. A., & Glutting, J. J. (2012). Knowledge,
writing, and language outcomes for a reading comprehension and writing
intervention. Reading and Writing, 26(7), 1-26.
Morphy, P., & Graham, S. (2012). Word processing programs and weaker writers/readers: A meta-
analysis of research findings. Reading and Writing, 25, 641-678.
National Commission on Writing (2003). The need for a writing revolution. The neglected “R”.
New York: College Entrance Examination Board.
Northwest Regional Educational Library (2013). About 6 + 1 Trait Writing. Retrieved September 2,
2013, from http://www.educationnorthwest.og/resource/949.
*Olson, V. L. B. (1990). The revising processes of sixth-grade writers with and without peer
feedback. The Journal of Educational Research, 84, 22-29.
*Paquette, K. R. (2008). Integrating the 6+ 1 Writing Traits Model with Cross-Age Tutoring: An
Investigation of Elementary Students' Writing Development. Literacy Research and
Instruction, 48(1), 28-38.
*Puma, M., Tarkow, A., & Puma, A. (2007). The Challenge of Improving Children's Writing
Ability: A Randomized Evaluation of" Writing Wings". Institute of Education Sciences.
Retrieved September 2, 2013 from:
*Raphael, T. E., & Kirschner, B. M. (1985). The Effects of Instruction in Compare/Contrast Text
Structure on Sixth-Grade Students' Reading Comprehension and Writing Products. Research
Series No. 161.
Rijlaarsdam. G., Bergh, H. van den, Couzijn, M. Janssen, T., Braaksma, M., Tillema, M. Steendam,
E. van, Raedts, M. (2011). Writing. In, Graham, S., Bus, A. Major, S. & Swanson, L. (Eds.),
Application of Educational Psychology to Learning and Teaching. APA Handbook Volume 3,
Rogers, L., & Graham, S. (2008). A meta-analysis of single subject design writing intervention
research. Journal of Educational Psychology, 100(4), 879-906.
*Ross, J. A., Rolheiser, C., & Hogaboam-Gray, A. (1999). Effects of self-evaluation training on
narrative writing. Assessing Writing, 6(1), 107-132.
*Saddler, B., & Graham, S. (2005). The effects of peer-assisted sentence-combining instruction on
the writing performance of more and less skilled young writers. Journal of Educational
Psychology, 97(1), 43-54.
Salahu-Din, D., Persky, H., & Miller, J. (2008). The Nation’s Report Card: Writing 2007 (NCES
2008-468). National Center for Education Statistics, Institute of Education Sciences, U.S.
Department of Education, Washington, D.C.
Schoonen, R. (2005). Generalizability of writing scores: An application of structural equation
modelling. Language Testing, 22, 1-30.
*Schunk, D. H., & Swartz, C. W. (1993). Goals and progress feedback: Effects on self-efficacy and
writing achievement. Contemporary Educational Psychology, 18(3), 337-354.
Smits, M. (2009). Schrijven en leren op de pabo. [Writing and learning in teacher training] Doctoral
dissertation. Nijmegen: Radboud University.
Stoeldraijer, J. (2012). Kwaliteitskaart onderwijs in het schrijven van teksten.[Quality card
education in writing texts] The Hague: School aan Zet.
*Tienken, C. H., & Achilles, C. M. (2003). Changing Teacher Behavior and Improving Student
Writing Achievement. Planning and Changing, 34(3), 153-168.
*Torrance, M., Fidalgo, R., & García, J. N. (2007). The teachability and effectiveness of cognitive
self-regulation in sixth-grade writers. Learning and Instruction, 17(3), 265-285.
*Varble, M. E. (1990). Analysis of writing samples of students taught by teachers using whole
language and traditional approaches. The Journal of Educational Research, 245-251.
, B
. Y., Hoskyn,
aches to teac
, 33(4), 7
F., & Toppin
tructured peer
: Forest plot
er study
., Jai, D., Ellis,
7-784. http://d
. J. (2001). Col
interaction. Bri
f expected (
P., & Watson,
ers opinion
aborative writi
ish Journal of
rey) and obse
. (2008). The c
ssay writin
: The effects
Educational Ps
rved effect si
mparative effi
f metaco
, 71(2)
e, and 95% c
acy of two
, 261-282.
... Findings of studies from different countries suggest that a very large percentage of students are experiencing difficulties with their writing skills (National Center for Education Statistics, 2012; Alves and Limpo, 2015;Koster et al., 2015). However, as Graham and Perin (2007) highlighted, not all the children that experience writing difficulties are identified as having a learning disability; low-achieving writers are included in this percentage of students with writing difficulties as a silent majority who lack writing proficiency but do not receive additional help. ...
... In a similar way, Vander Hart et al. (2010) study suggested that even though kindergarten teachers employ several effective intervention strategies for writing, there is room for improvement on implementing good handwriting practices based on research. In addition, studies by Gilbert and Graham (2010) and Koster et al. (2015) pointed out that the time devoted to writing in elementary schools is limited, and only a minority of schools and teachers used evidenced-based instructional practices. ...
... Promoting writing fluency with evidence-based materials is particularly important in the initial years of learning to write. Nevertheless, previous reviews and meta-analyses on evidencebased writing interventions have focused mainly on children and youngsters in first grade and beyond; therefore, studies involving kindergartens were excluded and writing fluency-based outcomes were missing in most of the writing intervention reviews and meta-analyses done to date (Rogers and Graham, 2008;Graham and Sandmel, 2011;Graham et al., 2012Graham et al., , 2015Koster et al., 2015;. In this sense, it is worth mentioning review, in which the authors conducted an extensive synthesis on writing instruction in elementary grades. ...
Full-text available
Handwriting is a complex activity that involves continuous interaction between lower-level handwriting and motor skills and higher-order cognitive processes. It is important to allocate mental resources to these high-order processes since these processes place a great demand on cognitive capacity. This is possible when lower-level skills such as transcription are effortlessness and fluent. Given that fluency is a value in virtually all areas of academic learning, schools should provide instructional activities to promote writing fluency from the first stages of learning to write. In an effort to determine if teaching handwriting enhances writing fluency, we conducted a systematic and meta-analytic review of the writing fluency intervention literature. We selected 31 studies: 21 true and quasi-experimental studies, 4 single-group design, 3 single-subject design, and 3 non-experimental studies, conducted with K-6 students in a regular school setting. A total of 2,030 students participated in these studies. When compared to no instruction or non-handwriting instructional conditions, teaching different handwriting intervention programs resulted in statistically significant greater writing fluency (ES = 0.64). Moreover, three specific handwriting interventions yielded statistically significant results in improving writing fluency, when compared to other handwriting interventions or to typical handwriting instruction conditions: handwriting focused on training timed transcription skills (ES = 0.49), multicomponent handwriting treatments (ES = 0.40), and performance feedback (ES = 0.36). There were not enough data to calculate the impact of sensory-motor and self-regulated strategy handwriting interventions on writing fluency. The significance of these findings for implementing and differentiating handwriting fluency instruction and guiding future research will be discussed.
... In view of these needs, efforts have been made to develop effective interventions, and international research related to instructional practices in the writing domain has been consensual in demonstrating the success of instructional models for writing that focus on both the explicit instruction of mechanical and content-related writing strategies along with the procedures for regulating those strategies (Graham & Perin, 2007;Graham et al., 2012;Koster et al., 2015). More specifically, the Self-Regulated Strategy Development model (SRSD; e.g., Mason et al., 2011) emerges as one of the most effective evidencebased methods with promising results. ...
... Together, the explicit instruction of self-regulation strategies, the use of engaging methodologies and the support of a narrative that models and guides the training, CriaTivo enables the development of transversal learning-related competencies (e.g., metacognitive strategies) while applying them to the writing process (e.g., planning and monitoring the written composition, time management, self-reflection on the process and the writing product). Therefore, CriaTivo appears to integrate a set of promising tools to improve students' writing performance Koster et al., 2015). Thus, in order to assess the effectiveness of CriaTivo, two hypotheses were established: H1: At posttest, elementary-school students will increase their use of self-regulatory strategies associated with the writing process, namely Forethought/Planning, Monitoring and Revising skills, compared to pretest. ...
Full-text available
Writing has a leading role in learning and, although elementary-school curricula emphasize the development of this complex skill, many students still struggle with their writing performance. This study aimed to assess the efficacy and social validity of CriaTivo, a curriculum-based intervention developed following a Response to Intervention model to promote self-regulation of the writing process (i.e., planning, monitoring, revising) applied to the written composition of narrative texts across third and fourth grades. Two hundred eighty-one Portuguese students (55% boys, M = 8.58 years, SD = 0.79) and their teachers participated in the study. A mixed-methods research design was used, and data was collected at two points in time. Regarding the intervention’s efficacy, results were promising, depicting improvements at posttest in students’ planning and monitoring skills, as also in their writing quality. The findings also supported the intervention’s social validity for both students and teachers. Despite requiring further research, CriaTivo appears to be a promising curriculum-based intervention which responds to the previously identified research and practice needs.
... For example, there have also been meta-analyses for language teaching. Thus, Koster et al. (2015) and van Weijen and Janssen (2018) identified fective practices just listed: (1) goal setting; ...
... This goal-setting step at the beginning of the work as well as in the intermediate phases can take two different forms (KOSTER et al., 2015). The teacher can actually announce to the students the expected goals in terms of the product, for example, the overall length of the text, the expected effects, and so forth. ...
Full-text available
ABSTRACT This contribution has a twofold objective: (1) to list effective practices for teaching oral language through genres, based on the latest research findings and (2) to propose a training program that uses these practices. To achieve this aim, we first explore effective practices for teaching oral language through genres, based on the literature. In the second part, we present Itineraries, an instructional program that implements these effective practices. Cette contribution a un double objectif : (1) recenser les pratiques effica-ces pour l'enseignement de la compétence à communiquer oralement par les genres, sur la base des derniers travaux de recherche et (2) proposer un dispositif d'enseignement/apprentissage qui mobilise ces pratiques. Dans le texte, nous déployons la première partie les pratiques efficaces pour l'enseignement de l'oral par les genres en nous appuyant sur la litté-rature, nous permettant d'aboutir à un modèle théorique. Dans la deu-xième partie du texte, nous présentons le dispositif "Itinéraires" étape par étape et montrons comment il peut être utilisé dans les classes.
... Still, pretesting is not universally employed in writing research. A meta-analysis on writing intervention research in elementary schools from Koster et al. (2015) showed that 34% of studies used a posttest-only design. Furthermore, 47% used a pretest-posttest design with the posttest immediately following the intervention, instead of employing a delayed posttest, which Koster et al. deemed better suited to make claims about long-term effectiveness. ...
Full-text available
In popularization discourse, insights from academic discourse are recontextualized and reformulated into newsworthy, understandable knowledge for a lay audience. Training in popularization discourse is a relatively new and unexplored research topic. Existing studies in the science communication field suffer from under-utilized baseline assessments and pretests in teaching interventions. This methodological problem leads both to a lack of evidence for claims about student progress and to a gap in knowledge about baseline popularization skills. We draw the topic into the realm of writing research by conducting a baseline assessment of pre-training popularization skills in first-year undergraduate students. Undergraduate science communication texts are analyzed to identify instances of popularization strategies using a coding scheme for text analysis of popularization discourse. The results indicate a lack of genre knowledge in both academic and popularized discourse: textual styles are either too academic or overly popularized; the academic text is misrepresented; and the essential journalistic structure lacking. An educational program in popularization discourse should therefore focus on the genre demands of popularization discourse, awareness of academic writing conventions, the genre change between academic and popularized writing, the role of the student as a writer, and stylistic attributes.
... Sánchez-Rivero et al. (2021) conducted a questionnaire study to investigate whether elementary and secondary school teachers in Spain implemented evidence-based practices in their classroom. To identify the evidence-based practices to teach writing, they designed a questionnaire based on the five available meta-analyses (Graham & Harris, 2018;Graham et al., 2015;Graham et al., 2012;Graham & Perin, 2007;Koster et al., 2015) and found that Spanish elementary and secondary teachers did not very frequently use evidence-based practices (it was even less frequent in the case of secondary school teaching). As mentioned earlier, the use of evidence-based practices was connected to several other variables, including self-efficacy, attitudes, and preparedness. ...
Full-text available
This paper investigates the reported frequency of the use of evidence-based writing practices (EBWPs) by teachers (N = 51) in primary and secondary school classrooms in a sample of schools in the Barcelona metropolitan area (Spain), and how teacher beliefs contribute to the reported use of EBWPs. The results showed that the teachers declared to implement most of the EBWPs from previous studies. The three most frequent declared practices were 1) give praise individually for writing, 2) teaching writing strategies for planning and writing skills, and 3) using text assessment as a guide to shape instructions. Regarding teachers’ beliefs about teaching writing, the study focused on teachers’ attitudes and teacher efficacy. The results on attitude showed that teachers had a positive attitude toward writing. Results regarding teacher efficacy showed that teachers felt quite efficacious, especially when they were required to determine the level of difficulty in written assignments. A factor analysis of the EBWPs showed that the two main factors for the frequency of reported use of EBWPS were strategy teaching for evidence-based writing and writing practices based on text assessment. PLS regression analyses showed that the reported frequency of use of EBWPs was highly predicted by the feeling of efficacy of teachers.
... The study did not identify a moderating effect of duration of intervention on effect size estimates, either. This finding confirms a meta-analysis conducted by Koster et al. (2015), which noted that the effectiveness of writing interventions was not associated with the duration of intervention. Dignath et al. (2008) also contended that the length of intervention did not significantly impact the reading/writing performance. ...
Full-text available
Self-Regulated Strategy Development (SRSD) is a criterion-based teaching approach incorporating explicit and systematic instruction of writing strategies, knowledge or skills, and self-regulation procedures into writing. This study aims to estimate the overall average effect of SRSD on English writing outcomes and to examine the extent to which the effect was moderated by outcome features (writing outcome, genre, measure type), methodological features (assignment, duration of intervention, and methodological quality), learner features (educational level and country), and publication features (publication type). A total of 43 effect sizes were extracted from 22 primary studies that were published or written in the past decade. Results suggested that SRSD had a positively large effect on writing outcomes. Writing outcome type was found to moderate effect size estimates. SRSD had a statistically smaller effect on writing length than on quality and elements. In addition, published journal articles had significantly larger effect size estimates than dissertations. The current meta-analysis informs practitioners, classroom teachers, and program designers of the overall effect of SRSD on English writing and factors related to the effectiveness of SRSD interventions.
... A subsequent metaanalysis (Graham et al., 2015) found the average effect size for various types of feedback to be d = 0.61 and the effect of adult feedback to be d = 0.87. Similar findings were also reported by Koster et al. (2015). Koenka et al. (2019) even found a positive effect of providing feedback in the form of grades -as compared to no feedback at all -although written comments were more effective than grades, with an average effect size of d = 0.30 for comments versus grades. ...
... These interventions span word, sentence and text level aspects. Meta-analyses have found positive effects for sentence combining and negative effects for explicitly teaching grammar (Koster, Tribushinina, De Jong & Van den Bergh, 2015;Graham, et al., 2012; see also Graham & Perin, 2007). Sentence combining requires pupils to combine sentences in various ways, for example through being able to combine two grammatically simple sentences into a complex sentence using linking words such as because, when, etc. ...
Full-text available
Dans le cadre de cette contribution, nous avons choisi de centrer notre attention sur l’accompagnement de l’écriture d’un portfolio à double enjeu, destiné à des jeunes formateurs engagés dans l’enseignement supérieur et amenés à produire un texte réflexif qui se veut un levier de développement professionnel et constitue en même temps un support à l’évaluation certificative d’un programme d’agrégation de l’enseignement supérieur. Les balises pour accompagner les étudiants dans la production d'un portfolio sont présentés dans le texte.
Children come to school with an idea of narrative structure, but their knowledge of persuasive writing is less developed. Results of the National Assessment of Educational Progress indicate that students perform poorly on persuasive writing tasks. To what extent this is a result of a lack of logical thinking skills and to what extent it is a result of problems with instruction are not known. The present experiment was designed to identify the effects of instruction, grade level, and sex on students’ persuasive writing. Children in grades 4, 6, and 8 were instructed in writing with one of four instructional strategies. Their performance was then evaluated through analytic scoring of their efforts with standard writing prompts administered immediately after treatment and again 2 weeks later. The results showed that older children wrote better than younger ones; that girls wrote better than boys immediately after the study but not 2 weeks later; and that treatment effects approached statistical significance.
This study examined whether instruction in genre knowledge enriches students’ feedback on each other’s writing, resulting in better writing quality. In total 140 sixth-grade students (age 11–13) participated in the study. Two approaches to peer response with additional instruction were compared. In one condition, students were taught specific genre knowledge (SGK). In another condition, students were taught general aspects of communicative writing (GACW). Both groups were compared with a baseline control group. Students were randomly assigned to the conditions. Results showed strong effects of the SGK condition outperforming the other conditions on text quality of four posttest writing tasks. Video recordings of students commenting on each other’s first drafts showed that the students in the SGK condition gave significantly more attention to the functions taught than students in the GACW condition. This finding supports the interpretation that knowledge about the genre-specific functions was actually used toimprove texts.
The purpose of the study was to investigate whether instruction aimed to heighten awareness of narrative structure would enhance fifth-grade children's use of story elements, as one strategy, during comprehension and composition. Evidence on experimental and and control group strategy use was obtained by using dependent measures plus introspective means. No significant differences were found between the groups on any of the multiple dependent measures following five weeks of instruction. The results of the selfreports confirmed that instructional elements (e.g., self-questioning, monologuing) common to both treatment groups may have been key contributors to the no significant difference findings. Self-reports, however, provided further evidence that children use similar strategies in the reading and writing processes.
The primary purpose of this study was to examine the effect of peer feedback on the quality of student writing and the amount and kind of revision behavior. Ninety-three sixth graders in six intact classrooms wrote and revised six stories, the last one being used as data for this study. Instruction varied across groups in the following manner: RI/PP students received revision instruction and revised stories with a peer; PP students revised stories with a peer but did not receive revision instruction; RI students received revision instruction but revised stories alone; C students had neither revision instruction nor help from peers. Chi-square analysis indicated that revision behavior was influenced by instruction. Quality of writing analysis revealed significant differences across groups on both rough and final drafts. Peer feedback seemed to help students write initially superior rough drafts but was not consistently linked to improvement of content between rough and final drafts. Successful surface structure editing occurred with or without peer feedback.
The purpose of this study was to examine the quality of writing of second and sixth graders who were taught by two different writing approaches (whole language and traditional). The writing samples were rated on two separate criteria-quality of content and mastery of mechanics. Three raters evaluated the writing samples, and an analysis of variance was performed to measure differences. The findings were as follows: (a) Second graders who were taught by teachers using the whole language approach produced better writing samples when evaluated on meaning and content, (b) There was no difference in writing samples in the correct use of mechanics of second graders taught by either approach, (c) There was no difference in writing samples of sixth graders taught by either approach.