Journal of Educational Psychology

© 2023 American Psychological Association 2023, Vol. 115, No. 7, 1004–1027

ISSN: 0022-0663 https://doi.org/10.1037/edu0000819

A Meta-Analysis of Writing Treatments for Students in Grades 6–12

Steve Graham1, Young-Suk Kim2, Yucheng Cao2, Will Lee2, Tamara Tate2, Penelope Collins2, Minkyung Cho2,
Youngsun Moon2, Huy Quoc Chung2, and Carol Booth Olson2
1
School of Education, Mary Lou Fulton Teachers College, Arizona State University
2
School of Education, University of California, Irvine

There is considerable concern that many adolescents do not attain the writing competence needed to be suc-
cessful in school, their personal lives, or the workplace. Ensuring that students acquire this competence is a
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

basic responsibility of schools. In order to meet this objective, teachers need access to effective practices
This document is copyrighted by the American Psychological Association or one of its allied publishers.

for teaching writing. In this meta-analysis, we examined if teaching writing improved the writing and reading
of students in Grades 6–12, and what specific writing treatments enhanced students’ writing. Our review
included writing treatments tested using an experimental or quasi-experimental design (with pretests) and pub-
lished and unpublished studies, and computed effect sizes (ESs) for all writing and reading outcomes assessed.
Across 406 independent comparisons, yielding 3,514 ESs involving 52,529 students, teaching writing had a
positive and statistically detectable impact on students’ writing (ES = 0.47) and reading (ES = 0.22).
Moreover, a variety of different writing treatments improved students’ performance on writing measures.
Across all writing outcomes, statistically detectable effects (presented in parentheses) were obtained for com-
prehensive writing programs (0.47; which included the process approach to writing), strategy instruction
(0.76), digital writing tools (0.31), transcription instruction (0.71), computer-assisted instruction (0.32), teach-
ing critical/creative thinking skills for writing (0.27), emulating good models of writing (0.46), feedback
(0.34), goal setting (0.44), prewriting activities (0.49), grammar instruction (0.77), sentence instruction
(0.73), inquiry (0.92), observing writers/readers, peer assistance (0.41), summarization instruction (0.49),
and text structure instruction (0.39). Implications for practice, research, and theory are discussed.

Educational Impact and Implications Statement

In this meta-analysis, we identified instructional practices that improve the writing of students in Grades
6–12. We found that multiple procedures can improve these students’ writing including the process writ-
ing approach, setting writing goals, prewriting and inquiry activities for gathering and organizing writ-
ing content, emulating models of good writing, writing with digital tools, peers assisting each other with
writing, students observing other writers and readers interacting with text, as well as teaching writing
strategies, grammar, spelling, sentence construction, summary writing, creative or critical thinking
skills, and text structure instruction. Teaching writing also improved Grade 6–12 students’ reading.

Keywords: writing, composition, instruction, middle school, high school

Supplemental materials: https://doi.org/10.1037/edu0000819.supp

A basic responsibility of schools is to help students become com- starting in kindergarten (Tolchinsky, 2016), secondary schools
petent writers (National Commission on Writing, 2003). Although play an important role in students’ writing development. This
writing instruction begins early in the schooling process, often includes helping them extend the writing capabilities they developed

supporting role for conceptualization, data curation, and writing–review

Steve Graham https://orcid.org/0000-0002-6702-5865 and editing. Minkyung Cho served in a supporting role for conceptualiza-
The research reported here was supported by the Institute of Education tion and data curation. Youngsun Moon served in a supporting role for con-
Sciences, U.S. Department of Education, through Grant R305C190007 to ceptualization and data curation. Huy Quoc Chung served in a supporting
the University of California, Irvine, for the WRITE Center. The opinions role for writing–original draft. Carol Booth Olson served as lead for funding
expressed are those of the authors and do not represent views of the acquisition and served in a supporting role for writing–review and editing.
Institute or the U.S. Department of Education. Steve Graham and Young-Suk Kim contributed equally to data curation.
Steve Graham served as lead for conceptualization, supervision, writing– Steve Graham and Yucheng Cao contributed equally to methodology and
original draft, and writing–review and editing. Young-Suk Kim served in a formal analysis. Young-Suk Kim and Will Lee contributed equally to
supporting role for methodology and writing–review and editing. Yucheng conceptualization.
Cao served in a supporting role for conceptualization and writing–original Correspondence concerning this article should be addressed to Steve
draft. Will Lee served in a supporting role for data curation and visualiza- Graham, School of Education, Mary Lou Fulton Teachers College, Arizona
tion. Tamara Tate served in a supporting role for conceptualization, data State University, P.O. Box 871811, Tempe, AZ 85287, United States.
curation, and writing–review and editing. Penelope Collins served in a Email: [email protected]

1004
 SECONDARY WRITING INSTRUCTION 1005

in the elementary grades as well as expanding their capabilities to knowledge of different types of writing), and writing motivational
use writing as a tool for analyzing and learning content in subject- beliefs (e.g., efficacy, attitudes, and motives for writing). Outcomes
matter classes (Hillocks, 2008). The available evidence demon- for reading, in contrast, were more limited as studies only included
strates many schools are not meeting these goals. In the United measures of reading achievement (these were always norm-
States, for example, two-thirds of secondary students’ scores on referenced measures), reading comprehension, word reading, read-
the last national assessment were so low they exhibited only partial ing vocabulary, and reading motivation. It should be noted that read-
mastery of the writing capabilities needed for grade-level success ing achievement measures in this review were unstandardized or
(National Center for Educational Statistics, 2012). Unfortunately, standardized assessments that evaluated reading competence across
concerns about poor writing are not limited to the United States, multiple reading constructs (including reading comprehension in
but are common around the world (Graham & Rijlaarsdam, 2016). many instances).
A useful approach for identifying instructional practices that have
the potential to transform writing instruction and students’ writing at Theoretical Underpinnings
the secondary level is the systematic review of writing intervention
Writing
research. In this article, we report findings from a comprehensive
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

meta-analysis of experimental and quasi-experimental writing stud- The writer(s)-within-community model of writing (WWC;
ies conducted with students in Grades 6–12. The purpose of this Graham, 2018) served as the primary foundation for this meta-
meta-analysis was to identify effective practices for teaching writing analysis. The basic underlying tenet of the model is that writing
at the secondary level. Such a review is needed for the following rea- and the teaching of writing are simultaneously and interactively
sons. One, teachers need access to teaching practices with a track shaped by the context in which they occur and the cognitive capabil-
record of proven success if they are to provide better writing instruc- ities and resources of the writers in these contexts. Writing and the
tion. Studies conducted in many countries have raised concerns teaching of writing in school are determined and bound by its pur-
about the quality of writing instruction provided to secondary stu- poses in this context, including goals for writing, its value and
dents (e.g., Applebee & Langer, 2011; Drew et al., 2017). Two, a motives, audiences, established norms, social practices, writing sup-
meta-analysis such as the current one provides information that ports, writing stance/identity of the class, available tools for writing,
can inform writing theory (Graham & Harris, 2014). For instance, typical writing practices, types of writing products produced, expe-
if instructional practices designed to improve students’ sentence riences and resources of classroom members (teachers and students),
skills enhances writing quality, this provides evidence that sentence physical and social environment established, and the collective his-
skills are an important component of students’ writing. Three, meta- tory of the class. What students write and how writing is taught are
analyses that assess study quality, as was done in this review, can also shaped by macrofactors including institutional, social, cultural,
provide useful information on strengths and weaknesses of a body political, and historical factors operating outside the class.
of research, resulting in recommendations for improving future Writing and teaching writing within a classroom community are
research. accomplished by its members (teachers and students) and the cogni-
Another reason why the current review is needed is because the tive capabilities and resources they possess (Graham, 2018).
last comprehensive meta-analysis with secondary students was in According to the WWC model, students consciously and deliber-
2007 (Graham and Perin). The search for this prior review ended ately establish executive control over what they write by establishing
May 2005 and identified 94 comparisons with Grade 6–12 students their own writing goals (usually in concert with teacher and class-
that assessed the effects of writing treatments on the quality of stu- room goals). To achieve these desired goals, students activate,
dents’ writing (the only outcome measure examined). Our current orchestrate, and adjust as necessary relevant knowledge and beliefs
meta-analysis included 406 writing treatment/control comparisons held in long-term memory (LTM), useful tools (e.g., spell checkers)
with students in Grades 6–12 (more than a fourfold increase in com- and resources (e.g., information from an online site) available to the
parisons), and it examined the effects of teaching writing on all writ- writer within or outside the classroom, and the production processes
ing and reading outcomes. Writing treatments in the current needed to produce writing. These cognitive processes may be
meta-analysis included practice writing (e.g., increased writing); altered by emotional responses to writing, personality traits, and
procedures or tools that support students as they write (e.g., goal set- physiological states. These same cognitive resources and processes
ting, inquiry, prewriting activities, digital writing tools, feedback, (executive control mechanisms, LTM resources, production pro-
and peer assistance); teaching writing skills (e.g., spelling, grammar, cesses, and emotional, psychological, and physical modulators)
and sentence construction), processes (e.g., strategy instruction, revi- shape and constrain teachers’ actions when teaching writing.
sion instruction, imagery, and summarization instruction), and The control processes that students and teachers apply when writ-
knowledge (e.g., text structure instruction, vocabulary instruction, ing and teaching writing, respectively, include directing and sustain-
and emulating good models of writing); specific methods for acquir- ing attention, working memory, and executive control, which
ing increased writing competence (e.g., observation and computer- includes establishing goals for writing, devising plans or using pre-
assisted instruction); and comprehensive writing programs (e.g., existing schemas to accomplish them, and monitoring and adjusting
the process approach to writing). goals and plans as needed. LTM sources include specialized knowl-
In terms of outcomes, all measures of writing and reading were edge about how to write (e.g., knowledge of spelling; handwriting;
included in our meta-analysis. For writing, this included general typing; keyboarding; punctuation; capitalization; grammar and sen-
measures of performance (e.g., writing quality) as well as measures tence instruction; purposes and features of different types of text;
examining specific characteristics of text (e.g., ideation, organiza- qualities of good writing; schemas and strategies for planning, draft-
tion, and voice), writing skills (e.g., spelling, handwriting), writing ing, and revising text; and tools for writing) as well as knowledge
process (e.g., planning, revising), writing knowledge (e.g., genre about content for writing, intended audiences, oral language
 1006 GRAHAM ET AL.

(which forms the foundation for written communication), and read- bidirectional relations where development of one bootstrap the
ing (reading is used to gather writing information and to when eval- other through learning experiences. In addition, the reading–writing
uating and revising writing). LTM also includes a variety of beliefs connections also posited to be dynamic such that their relations are
that can both power and constrain students’ writing efforts, including influenced by several factors such as development and measurement.
efficacy, attitudes/interest in writing, value and utility placed on writ- According to these theories, writing instruction designed to
ing, motives for writing, reasons for writing success/failure, and stu- enhance any of these four pools of knowledge should improve stu-
dents’ identities as writers. Finally, the process that writers use to dents’ reading skills, knowledge, performance, and even motivation.
produce text include: conceptualization (goals, plans, and mental For example, writing instruction designed to increase students’
representation of the writing task), ideation (acquiring content for knowledge about the functions and purposes of text should result
writing from LTM or external sources), translation (turning pertinent in better reading because students can use such knowledge to help
ideas into acceptable phrases and sentences), transcription (tran- them better understand an author’s intentions. Similarly, increasing
scribing phrases and sentences into written text), and reconceptual- how much students write can introduces them to new content,
ization (rethinking and revising plans and text). which can provide them with a stronger foundation for understand-
The WWC model (Graham, 2018) helped us reduce the multitude ing text about the same content. Thus, we anticipated that teaching
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

of different writing outcomes included in the studies reviewed to a writing would have a positive impact on Grade 6–12 students’
more manageable set of 10 writing constructs (see Coding of reading.
Study Features section). For example, skills involved in sentence
construction (e.g., sentence fluency, syntax, vocabulary, and writing Previous Meta-Analyses
conventions) were grouped together to represent the WWC produc-
Writing
tion process of translation. Similarly, assessments of writing effi-
cacy, attitudes toward writing, and other beliefs about writing were The earliest meta-analysis to conduct a comprehensive review of
combined to represent the construct of writing motivational beliefs. writing treatments was Hillocks (1986). He examined the impact of
In addition, the WWC model provided guidance for categorizing dif- teaching writing on the quality of students’ writing in Grade 3
ferent writing treatments and a framework for considering what cog- through college, drawing on studies with experimental or
nitive processes or LTM resources an effective writing treatment quasi-experimental designs. He computed effect sizes (ESs) for 75
impacted (see Table 1). For instance, spelling, handwriting, and typ- treatment/control comparisons (adjusted for pretest differences)
ing instructional studies were categorized as transcription instruc- included in 60 documents. The ES for all comparisons across all
tion, but writing strategies instruction (if effective) was assumed to grades was 0.28. Separate ESs were not provided for students in
provide useful schemas for writing for executive control processes Grades 6–12 (our focus in this review).
and useful heuristics for the writing production processes of concep- The second comprehensive meta-analysis of writing treatments
tualization and reconceptualization. was published 21 years later (Graham & Perin, 2007). In this review,
154 treatment/control experimental or quasi-experimental compari-
Reading sons included in 123 documents with students in Grades 4–12 were
examined (all pertinent studies from Hillocks, 1986 were included in
The WWC model (Graham, 2018) views reading as an important this review). No overall ES for all studies across all grades was cal-
aspect of writing as writers use reading to obtain possible writing culated, quasi-experimental studies were included with or without
content as well as monitor and evaluate what they write. The pretests, and separate ESs were not computed for students in
model does not, however, describe how teaching writing may Grades 6–12. The statistically significant ESs for writing quality
improve reading. The shared knowledge hypothesis of reading and (in parentheses) for specific treatments were: strategy instruction
writing connections (Fitzgerald & Shanahan, 2000; Shanahan, (0.82), summarization instruction (0.82), sentence combining
2016) and the interactive dynamic literacy model (Kim, 2020, instruction (0.50), grammar instruction (−0.32), inquiry (0.32), pre-
2022) provide the theoretical underpinnings for how writing instruc- writing activities (0.32), study of models (0.25), peer assistance
tion can enhance reading outcomes. According to these theories, (0.75), goal setting (0.70), word processing (0.55), and process writ-
writing and reading are not identical skills, but the reading and writ- ing approach (0.32). In terms of strategy instruction, the self-
ing processes draw on the same knowledge and cognitive systems. regulated strategy development model (SRSD ) was more effective
According to the shared knowledge hypothesis (Fitzgerald & (1.17) than all other forms of writing strategy instruction combined
Shanahan, 2000), there are four sources of knowledge students (0.69), and the process approach to writing was most effective (0.46)
draw from when writing and reading. This includes general knowl- when it involved National Writing Project (NWP) participation.
edge that students use to help them produce or comprehend text; There are two other comprehensive meta-analyses of writing treat-
meta-knowledge about written language (e.g., knowledge of the pur- ments that included at least some students in Grades 6–12 (Graham
poses and functions of text) students use to construct their own mes- et al., 2022; Koster et al., 2015). The meta-analysts conducting these
sages and interpret what an author means; procedural knowledge two reviews just included studies with experimental and
(e.g., goal setting, visualizing, summarizing, and analyzing) stu- quasi-experimental designs, focused just on writing quality out-
dents use to create text and understand it; and pragmatic knowledge comes, and did not provide separate ESs for Grades 6–12. The
of text attributes (e.g., features of text, words, syntax, and usage) stu- first review by Koster et al. (2015) included students in Grades 4–
dents employ to decode/encode words and comprehend/construct 6 (32 studies). The following writing treatments were statistically
sentences or larger units of text. According to the interactive and significant: goal setting (2.03), strategy instruction (0.96), text struc-
dynamic literacy model (Kim, 2020, 2022), reading and writing ture instruction (0.76), peer assistance (0.59), and feedback (0.88).
draw on multiple layers of skills, and they have interactive/ Teaching students to self-evaluate their writing (0.43) and grammar
 SECONDARY WRITING INSTRUCTION 1007

Table 1
Definition for Coded Writing Treatments and Cognitive Processes From WWC Model Impacted (in Parentheses)
Writing treatment Definition
Comprehensive writing Writing programs for teaching writing broadly. Such approaches can involve reading instruction, but the teaching of writing must
programs occur greater than 50% of the time. Comprehensive writing programs include commercial writing programs as well as programs
based on the process approach to writing. The process approach involves (sometimes referred to as writers’ workshop)
extended opportunities for writing; writing for real audiences; engaging in cycles of planning, translating, and reviewing;
personal responsibility and ownership of writing projects; high levels of student interactions and creation of a supportive
writing environment; self-reflection and evaluation; personalized individual assistance and instruction; and, in some instances,
systematic instruction. If process writing instruction was associated with the National Writing Project, this was coded. Programs
that taught specific skills (e.g., spelling) or writing processes (strategies for planning and revising) were not categorized as
comprehensive programs (all writing production processes, executive control, motivational beliefs, writing knowledge).
Strategy instruction Involves explicitly teaching strategies for planning, revising, self-assessing, and/or editing text. Writing strategies range from
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

processes, such as semantic webs to strategies designed for specific types of writing, such as stories. Although graphic
organizers, mind maps, and so forth can serve as a strategy, an instructional approach is not identified as strategy instruction
This document is copyrighted by the American Psychological Association or one of its allied publishers.

unless the procedure is modeled and students practice applying it at least two times. If writing strategies were taught using the
self-regulated strategy development model (Harris & Graham, 1992), this was coded. This approach is a special form of
strategy instruction that emphasizes teaching students the knowledge, skills, and self-regulation procedures needed to use
target strategies successfully (conceptualization, reconceptualization, executive control).
Digital writing This involved the use of digital tools for writing, including word processing; word processing plus programs with additional
features such as planning or evaluation aids, digital game elements, and developing a context in which writing occurs, as
well as speech-to-text synthesis. To qualify as digital writing, students had to use these tools over multiple sessions
(transcription, conceptualization, reconceptualization, ideation, motivational beliefs).
Transcription instruction Included teaching spelling, handwriting, or keyboarding (transcription).
Computer-assisted instruction Using a computer to teach writing skills, processes, or knowledge (all writing production processes).
Teaching critical/creative Instruction is designed to teach students to think and write critically, or to think and write more creatively. Can involve teaching
thinking skills for writing any aspect of critical or creativity thinking as it applies to writing (conceptualization and reconceptualization).
Emulating models Examining one or more examples of model texts or models for carrying out writing processes and attempting to emulate these
models when writing (writing knowledge).
Feedback Students received feedback about the adequacy of their text or progress in learning writing skills or processes. Feedback was
further coded as originating from teachers, peers, computers, or self (reconceptualization).
Goal setting Teacher assigned or students’ goals for writing or learning writing skills and processes (conceptualization, executive control).
Prewriting activities Use of activities such as using semantic webs, mind maps, brainstorming, searching the web for ideas, reading text for ideas,
dialoguing with others designed to help students generate or organize ideas for writing (ideation).
Procedural facilitation External supports (such as prompts, guides, hints, or heuristics) designed to enhance one or more writing processes, such as
planning or revising. These supports do not specifically focus on what students’ say when writing, but ensure specific writing
processes occur in an organized manner (conceptualization, reconceptualization, executive control).
Grammar instruction Explicit teaching of grammar (translation).
Sentence instruction Teaching students sentence construction (translation).
Imagery Using imagery as a way to enhance writing or as a method to learn specific writing skills or processes (ideation).
Inquiry Engaging students in activities to help them develop ideas for a particular writing task by analyzing immediate and concrete
data (e.g., comparing and contrasting cases or collecting and evaluating evidence) (ideation).
Enhancing motivation Writing instruction that includes multiple procedures for enhancing writing motivation (e.g., efficacy, attitudes, intrinsic, goal
orientation). Improving motivation is the stated purpose of the program (motivational beliefs).
Observation Observing other writers, readers of writing, or teachers/peers modeling how to carry out a writing process or skill (all writing
production processes).
Peer assistance Students working together to plan, draft, edit, and/or revise their compositions as well as learning via peer-tutoring, cooperative
learning, or students working together (all writing production processes).
Revising instruction Explicitly teaching students how to revise and edit text. Teaching students an editing or revision strategy via modeling and 2
days of practice was coded as strategy instruction (reconceptualization).
Summarization instruction Explicitly teaching students how to create written summaries (translation, conceptualization, reconceptualization, writing
knowledge).
Text structure instruction Explicitly teaching students knowledge about the purpose and/or structure of specific types of text, such as stories or persuasive
essays (writing knowledge).
Vocabulary instruction Teaching students vocabulary to use in their writing or vocabulary about their writing (e.g., red herring for mystery writing)
(translation).
Increased writing Increasing how much students write (all writing production processes).
Note. WWC = writer(s)-within-community model of writing.
 1008 GRAHAM ET AL.

instruction (−0.37) did not produce statistically detectable out- comparisons), reading fluency in Grades 1–7 (0.66; five compari-
comes. The second review by Graham et al. (2022) only involved sons), and reading comprehension in Grades 4–12 (0.22–0.27; 17
investigations conducted in Turkey with students in Grades 1 comparisons).
through college (77 studies). The ES for all experiments was 1.39,
and statistically significant ESs were obtained for strategy instruction
How the Current Meta-Analysis Replicated and Extended
(1.28), peer assistance (0.92), prewriting (1.55), and the process
writing approach (1.30). The especially large ESs obtained in this Previous Reviews
second meta-analysis may have been due to relatively weak control The review reported in the present article draws on and extends
comparisons. The control condition in all studies was previous meta-analyses of writing treatments in multiple ways.
business-as-usual (BAU), and writing instruction in Turkish schools Like most previous reviews, we focused on writing treatments
is often viewed as poor (e.g., Elma & Bütün, 2015). tested via experimental and quasi-experimental designs (with pre-
Other meta-analysts have examined the effects of specific writing test adjustments). We included both published and unpublished
treatments across elementary and secondary grades. For example, studies, which was also the case in the 20 meta-analyses used by
this includes meta-analyses of the effects of feedback (Graham
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Graham and Harris (2018). All eligible studies conducted with stu-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

et al., 2015), word processing (Goldberg et al., 2003), and grammar/ dents in Grades 6–12 from prior quantitative reviews of writing
sentence combining (Andrews et al., 2006). Drawing on both com- treatments were included in this meta-analysis. Moderator analyses
prehensive and treatment specific reviews (20 different quantitative were conducted to determine if study attributes accounted for
analyses in total), Graham and Harris (2018) conducted a meta- excess variance in ESs. This was done in most, but not all prior
analysis of writing treatments (266 comparisons) tested via experi- meta-analyses.
mental and quasi-experimental designs with students in Grades 1– In contrast to prior reviews, we exclusively focused on students in
12 (separate ESs were not reported for Grades 6–12). The following Grades 6–12. The number of studies reviewed was increased signifi-
writing treatments statistically improved writing quality: strategy cantly by conducting a new search of the literature, but more impor-
instruction (1.26), creativity/imagery instruction (0.76), emulating tantly, by including any study with one or more writing measures
model text (0.30), sentence combining instruction (0.50), teaching that assessed written products, writing skills, writing processes, writ-
transcription skills (0.55), process approach to writing (0.34), ing beliefs, and writing knowledge. Previous reviews examined
word processing (0.44), word processing plus programs (1.46), writing quality, and in some instances a few other writing outcomes.
peer assistance (0.74), goal setting (0.80), prewriting activities We applied a more expansive approach by including a broad array of
(0.55), inquiry (0.37), adult feedback (0.87), peer feedback (0.77), writing outcomes. Lastly, we examined if writing and teaching writ-
self-assessment (0.51), feedback from a computer (0.34), and extra ing impacted reading as well as writing
writing (0.24). Writing treatments that did not yield statistically
detectable effects were teaching text structure (0.30), grammar Research Question 1: Does teaching writing improve writing
instruction (−0.17), procedural facilitation (0.52), and 6 + 1 traits performance when all outcomes are considered collectively as
(0.05). Furthermore, SRSD produced a larger ES (1.59) than general well as when the focus is on specific outcomes of writing qual-
strategy instruction (0.56). ity, ideation, elements, organization, voice, writing process,
Most, but not all (Hillocks, 1986), of the available comprehensive translation skills, transcription skills, writing knowledge, and
meta-analyses of writing treatments included moderator analyses to writing motivational beliefs?
determine if specific study features were related to magnitude of ESs.
Typically, this was not the case for study quality (Graham et al., Research Question 2: Are effects for writing outcomes moder-
2022; Koster et al., 2015), grade level (Graham et al., 2022; ated by the quality of studies and specific study features?
Graham & Perin, 2007; Koster et al., 2015), type of students Research Question 3: What specific writing treatments improve
(Graham & Perin, 2007), duration of treatment, type of control con- students’ writing?
dition, or type of assessment (Koster et al., 2015). There were several
exceptions, however, as younger process-writing approach students Research Question 4: Are effects for specific writing treatments
made greater gains in writing quality than older ones (Graham & moderated by type of writing outcome, grade level of students,
Harris, 2018; Graham & Perin, 2007), and less capable writers and assessment timing (posttest vs. maintenance)?
made greater gains than stronger writers when provided strategy
instruction (Graham & Perin, 2007). Research Question 5: Does teaching writing improve reading
outcomes collectively, reading achievement, reading compre-
hension, and reading vocabulary?
Reading
Research Question 6: Are effects for reading outcomes moder-
Only one comprehensive meta-analysis has been conducted ated by the quality of studies and specific study features?
assessing the effects of writing treatments on reading outcomes.
Graham and Hebert (2011) examined if teaching writing enhanced To answer Research Question 1, we first computed an ES for all
the reading of students in Grades 1–12. All comparisons applied writing outcomes collectively. In almost all past meta-analyses of
an experimental or quasi-experimental design (with pretests). writing treatments, a single writing outcome was calculated for
Increasing how much students wrote (nine comparisons) resulted each writing treatment/control comparison (e.g., writing quality in
in a statistically significant ES of 0.35 for reading comprehension Hillocks, 1986) or ESs for all measures for each comparison were
for students in Grades 1–6. Teaching writing had positive and statis- averaged together to create a single outcome (Graham & Hebert,
tically significant effects on word reading for Grades 1–5 (0.62; six 2011). This was done in these prior reviews to avoid violating the
 SECONDARY WRITING INSTRUCTION 1009

assumption of independent data points fundamental to the statistical (Bloom et al., 1956); (b) typically developing students tend to
analyses used in these reviews. In the current meta-analysis, a now make greater academic progress during a year than less capable
widely used statistical procedure, robust variance estimation ones (Swanson et al., 2013); (c) students who write as they learn
(RVE) was applied. This technique is a method for including multi- new writing skills, knowledges, or processes may make greater pro-
ple and dependent ESs in a meta-analysis (Tipton, 2013). RVE uses gress than those who do not as the former has the opportunity to
a working model, which approximates the assumed dependence apply what is learned (Troia, 2014); (d) posttest outcomes tend to
between ES estimates even when knowledge of the exact depen- be higher than maintenance outcomes as instructional effects can
dence structure is not known. ESs, using RVE, were also calculated decline over time (Suggate, 2016); and (e) magnitude of effects
for 10 writing constructs identified earlier from the WWC model for a treatment may differ depending on the type of control condition
(Graham, 2018; e.g., writing quality, writing output, writing process, to which it is compared (Pressley et al., 2006). Again, such effects
and writing knowledge). It is not only important to determine the were not common in prior meta-analyses of writing interventions,
impact of writing treatments overall, but on specific aspects of writ- but it is important to test if these study features indicate possible
ing as well. This can pinpoint which aspects of writing are impacted biases in obtained outcomes.
by writing instruction. A third moderator analysis involved a single predictor, time spent
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

For Research Question 2, we conducted three separate moderator teaching writing. This variable was not included in the second mod-
analyses. One moderator analysis examined if the magnitude of erator analysis described above because it was only collected in 63%
effects was related to study quality indicators (research design, pub- of the writing treatment/control comparisons. Its inclusion in the sec-
lication type, treatment fidelity, N of 1 study [i.e., either all treatment ond moderator analyses would have resulted in the exclusion of a
or control students were taught together by a single instructor], large number of writing treatment/control comparisons. We were
teacher effects, measurement reliability, norm-referenced test, and particularly interested in time spent teaching writing because greater
attrition rate). We selected these study quality indicators for the fol- opportunities to learn should produce larger effects (Carroll, 1963).
lowing reasons. For Research Question 3, we computed an ES for 23 writing treat-
One, quasi-experiments may produce larger effects than experi- ments (e.g., prewriting activities, computer-assisted instruction,
ments because selective factors may work in favor of the former grammar instruction, and inquiry). These writing treatments
(Cheung & Slavin, 2016). Two, journal articles may evidence larger included writing instructional procedures identified in previous
effects than unpublished studies because reviewers are more likely to meta-analyses (Hillocks, 1986; e.g., grammar instruction) as well
recommend the publication of investigations that produce more pos- as other writing treatments (e.g., observation of writers and readers;
itive outcomes (Hopewell et al., 2009). Three, variability in effects Couzijn, 1999) that could potentially enhance one or more of the
may be related to whether investigators establish treatment fidelity writing outcomes specified earlier from the WWC model
because if fidelity is not established, the writing treatment may not (Graham, 2018). An ES for a writing treatment was only calculated
be fully implemented, mitigating possible writing effects (Borrelli, if there were at least four or more independent treatment/control
2011). Four, in studies that are N of 1, treatment, control, or both con- comparisons (a criterion common in most prior writing meta-
ditions include just one instructor teaching all students together. This analyses; Graham & Perin, 2007). Findings from previous meta-
increases the likelihood that obtained differences in writing or read- analyses demonstrated that the effects of writing treatments vary
ing outcomes are due to teacher effects (e.g., one condition has a par- (e.g., Graham & Harris, 2018; Hillocks, 1986; Koster et al., 2015),
ticularly motivating teacher and the other does not) and not treatment but these prior reviews were limited as they typically involved a sin-
effects (Cheung & Slavin, 2016). Five, magnitude of effects may be gle outcome (e.g., writing quality) and included many fewer treat-
related to who provides instruction (teacher effects) because some ment/control comparisons than the 406 comparisons in this
teachers are more effective than others (Nye et al., 2004). Six, vari- meta-analysis.
ability in effects may be related to whether an assessment is reliable For five of the 23 writing treatments, we computed separate ESs
or not because outcomes from administered assessments may be for specific approaches. For example, similar to Graham and Perin
inconsistent (Nunnally, 2017). Seven, writing effects from norm- (2007), two separate ESs were computed for writing strategy instruc-
referenced tests may be smaller than those from researcher-designed tion: SRSD and general strategy instruction. In contrast to other
assessments because the latter may be over aligned with the writing approaches to writing strategy instruction, SRSD also teaches stu-
treatment (Slavin & Madden, 2011). Eight, sample attrition over dents self-regulation procedures for regulating learned strategies,
20% can change the outcomes of a study (Schulz & Grimes, the writing process, and writing behaviors (Harris & Graham,
2002). Even though moderator analyses of study quality indicators 1992). Likewise, as done by Graham et al. (2015), separate ESs
have rarely predicted magnitude of effects in previous meta-analyses were computed for four types of feedback: teacher, peer, self, and
of writing treatments, we still examined them in this review because computer. Two separate ESs were calculated for transcription
they can indicate possible biases in findings (Cheung & Slavin, instruction, teaching spelling, and teaching handwriting. Three sep-
2016). arate ESs were computed for comprehensive writing programs: com-
A second moderator analysis focused on whether the magnitude prehensive writing programs that were not a process approach,
of ESs for all writing outcomes collectively differed by specific process writing approach involving the NWP, process writing
study features (grade level, typical vs. less capable writers, text approach not involving the NWP (see Graham & Perin, 2007). We
was composed during the writing treatment, assessment timing specifically calculated effects for process writing approaches involv-
[posttest vs. maintenance], and type of control condition). These ing the NWP because this is the largest professional organization for
study features were selected as moderators because (a) younger stu- teachers of writing, and it devotes considerable effort to professional
dents may generate larger effects than older students as the former development for implementing process writing approaches.
tend to make greater academic gains in a year than the latter Additionally, two separate ESs were calculated for writing
 1010 GRAHAM ET AL.

treatments that involved digital writing tools: word processing and hard of hearing, emotional and behavioral difficulties, and blind-
word processing plus programs (see Morphy & Graham, 2012). ness). Although we believe that writing instruction is important
To answer Research Question 4, two different moderator analyses for these students, the aim of this review was to draw recommenda-
were conducted. One moderator analysis examined if there were stat- tions for teaching writing to students within general education set-
istically detectable differences in the obtained ESs for the different tings. We also excluded studies if there were less than 10 students
writing outcomes for a specific writing treatment. The magnitude in the writing treatment or control condition. Such studies were
of effects for writing outcomes should vary depending on attributes excluded because investigations with a small number of students
of writing each treatment is designed to improve. As writing quality have a larger margin of error than studies with larger sample sizes.
was the most common outcome in prior meta-analyses, our analyses Investigations were excluded if they assessed differences in how a
were contrast coded so other writing outcomes were compared to writing assessment was administered (e.g., changes in writing direc-
writing quality. The second moderator analysis examined if grade tions). Studies were not included if they involved a single compari-
level and assessment timing (posttest vs. maintenance) were related son of writing via word processor or by hand (any included study
to magnitude of ESs. As noted above, students in earlier grades may comparing mode of composing had to provide students with practice
evidence greater effects than older students because younger stu- over time using the targeted writing tools). Studies were further
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

dents tend to make greater academic gains in a year than older excluded if they tested writing treatments using single participant
ones (Bloom et al., 1956). Additionally, posttest instructional out- designs, students served as their own controls, and regression dis-
comes tend to decline over time (Suggate, 2016). continuity designs. Finally, ex post facto designs (i.e., no actual
To determine if teaching writing improves reading (Research experimental manipulation was undertaken) were excluded, as
Question 5), we computed an average-weighted ESs for all reading were studies if students in the treatment and control conditions
outcomes collectively. Separate average-weighted ESs were also were from different school years (e.g., treatment provided in one
computed for reading achievement, reading comprehension, and year and control students were from classes in preceding or subse-
reading vocabulary. Other aspects of reading (e.g., reading motiva- quent school years). We did not assume that claims that students
tion) were not tested frequently enough to be included in our analy- were randomly assigned to classes by school administrators before
ses. We anticipated that teaching writing would positively impact the school year began constituted random assignment. Our experi-
reading outcomes. According to the shared knowledge of reading ence is that adjustments are made in at least some of these assign-
and writing connections hypothesis (Fitzgerald & Shanahan, ments as a result of interactions between principals, teachers, and
2000) and the interactive dynamic literacy model (Kim, 2020, parents. As a result, such a study had to have corresponding pretests
2022), reading and writing draw upon common cognitive resources for posttested writing measures.
and instructional efforts to improve either one of these skills should
result in some improvement in the other. Furthermore, a prior meta- Search Procedures and Selection of Studies
analysis by Graham and Hebert (2011) found that teaching writing
improved the writing of students in Grades 1–12. We applied the following strategies to identify studies for this
To examine if magnitude of reading effects were related to study meta-analysis. One, electronic searches (ending September 30,
attributes (Research Question 6), we conducted the same moderator 2021) were conducted in the following databases: ERIC, APA
analyses that was applied to writing outcome collectively. The only PsycInfo, Linguistic and Language Behavior Abstracts, and
difference was the focus shifted to whether the moderators described ProQuest Dissertations & Theses Global. These searches involved
above for study quality and study features statistically predicted var- the following search terms: (writing) AND (instruct* OR treat*
iability in ESs for reading outcomes. The rationales for examining OR teach* OR handwrit* OR spell* OR grammar OR vocabulary
these specific moderators were identical to those specified for writ- OR “strategy instruction” OR “planning instruction” OR “revising
ing outcomes. instruction” OR sentence-combin* OR “sentence instruction” OR
“process approach” OR “computer-assisted instruct*” OR “word
Method process*” OR “writ* workshop” OR “national writ* project” OR
“peer collaborat*” OR “collaborat* writ*” OR “automated essay”
Inclusion Criteria
OR “goal set*” OR model OR “self-regulated strategy development”
Studies included in this review met the following criteria: (a) OR feedback OR usage OR mechanics OR “word processor” OR
Students were in Grades 6–12 (or ages 11–18); (b) students attended technology OR “on-line” OR “motivation and instruction” OR
regular public or private schools; (c) students in the treatment condi- “free writing” OR “additional writing” OR “extra writing” OR
tion wrote text, were taught writing, or both (50% or more of instruc- “peer-planning” OR “peer-revising” OR “self-evaluation” OR “self-
tional time was devoted to teaching writing); (d) an experimental or monitoring” OR “evaluative scales” OR “automated writing evalu-
quasi-experimental design was used to test the writing treatment ation” OR “mentor text” OR “formative assessment”) AND
(quasi-experiments had to have pretest measures for any writing out- (“Empirical Study” OR “Quantitative Study” OR “Longitudinal
come included in the analyses); (e) at least one aspect of writing was Study” OR “Follow up Study” OR “Experimental Replication”
assessed (this included measures of written products, writing skills, OR “true experiment*” OR “quasi experiment*” OR “randomized
writing processes, writing motivational beliefs, and knowledge per- control” OR “randomized cluster”).
tinent to writing); (f) the study was reported in English; and (g) data Two, we hand searched the following 16 journals from first date of
needed to calculate a weighted ES were provided or available from publication to September 30, 2021: American Educational Research
authors. Journal, Assessing Writing, British Educational Research Journal,
We excluded studies if students attended special schools or resi- British Journal of Educational Psychology, Journal of
dential settings for students with disabilities (e.g., deafness and Educational Psychology, Journal of Educational Research,
 SECONDARY WRITING INSTRUCTION 1011

Journal of Literacy Research, Journal of Research in Reading, students with special needs]), students composed text when writing
Journal of Research in Teaching English, Journal of Research on (yes [1]; no [0]), assessment timing (posttest [0]; an assessment
Educational Effectiveness, Journal of Writing Research, Learning delivered 2 or more weeks after posttest [1], and type of control con-
& Instruction, Reading & Writing, Reading Research Quarterly, dition (no instruction [1; control students received no writing treat-
Scientific Studies of Reading, and Written Communication. We ment], BAU [2; typical writing instruction provided to controls],
searched these journals as they commonly publish writing studies. diluted instruction [3; the control condition received part but not
Three, we consulted 24 previous meta-analyses focused on the all of the writing treatment], reading instruction [4; students in the
effectiveness of writing instruction (e.g., Graham et al., 2015; control condition read or were taught reading], alternative instruc-
Graham & Perin, 2007; Hillocks, 1986). Four, we contacted 35 tion [5; students in the control condition received a nonwriting or
authors who previously conducted writing treatment studies (e.g., nonreading treatment; e.g., math instruction], and counter [6; the
Rui Alves, Scott Crossley, Raquel Fidalgo, Teresa Limpo, Charles treatment and control conditions were virtually the same but the con-
MacArthur, Danielle MacNamara, Linda Mason, Deborah Myhill, trol condition differed in a significant way; e.g., word processing vs.
George Newell, and Gert Rijlaarsdam), asking them to share rele- handwriting text, peer vs. individual composing, individual vs group
vant studies or citations, including ones conducted by themselves instruction, computer-assisted vs. teacher instruction]). Student
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

or others. Five, all references cited in included documents were grade level was scored as the grade in which students were enrolled
searched to identify other pertinent studies. in during the time of the study. Time spent teaching writing was
The search process is depicted in Figure 1. After eliminating doc- coded as a continuous variable: hours of writing instruction
uments that did not meet inclusion criteria, 357 documents provided provided.
the data for this review. All titles and abstracts as well as all full doc- All writing measures in a study were named and described as
uments were examined to determine if they met inclusion/exclusion well, and coded as one of the following 10 outcomes derived ear-
criteria by a member of the research team. Twenty percent of lier from the WWC model (Graham, 2018): writing quality (e.g.,
these titles and abstracts and 25% of the full documents were holistic writing scale, analytic writing scale, or total score for
independently examined by a second team member. Agreement norm-referenced writing test), ideation (e.g., number of words,
for these determinations were 98% and 96% for the initial and full- sentence, idea-units, or T-units written as well as clarity, sub-
text screening, respectively. Differences were resolved through stance, and/or richness of ideas conveyed in a written text),
discussion. genre elements (e.g., number of schematic elements included in
a text, such as the inclusion of characters, setting, and events
Coding of Study Features in narrative writing), organization (e.g., logical progression of
ideas in a paper, including the establishment of transitions between
All studies were coded for the following four descriptive features paragraphs and sentences as well as coherence), voice, writing
of students: gender (male, female), race (Asian, Black, Hispanic, process (included planning, revising, perseverance, and self-
Native, White, and Other), language status (native language learners, regulation measures), translation skills (included sentence fluency,
second language learners, or a mixture of these two groups of stu- syntax, grammar, vocabulary, and convention measures), tran-
dents), socioeconomic status (SES) of students. SES was determined scription skills (included spelling, handwriting, and keyboarding
by the percent of students receiving free or reduced-priced lunch. measures), knowledge of writing (e.g., knowledge of writing
Five other descriptive features of studies included: study location topic, audience, genre, writing task, or conventions), and motiva-
(urban, suburban, rural), country (e.g., Finland), subject area (lan- tional beliefs (e.g., efficacy, attitudes, and motivation measures
guage arts, science, social studies, math, other), genre (narrative, such as intrinsic and extrinsic motivation).
expository, persuasive, multiple genre, other genres, no writing In 11 studies, an overall score for writing quality was not pro-
took place), and training for treatment implementers (yes or no). vided, but we were able to calculate one from multiple measures
Each study was also coded for the following eight categorical used in combination to assess writing quality: ideation, organization,
study quality indicators which served as moderators: research design voice, sentence fluency, vocabulary, and conventions. In such
(experiment [1] or quasi-experiment [0]), publication type ( journal instances, we averaged the outcomes for each of the attributes com-
article [1] or unpublished dissertation, conference paper, book/book monly used when assessing writing quality (e.g., ideation, organiza-
chapter, unpublished paper from author [0]), treatment fidelity tion, and sentence fluency) using the Nouri and Greenberg
(established through observation, teacher report, or researcher procedure (Cortina & Nouri, 2000) before computing an ES.
claim [1] or not established [0]), N of 1 investigation (not N of 1 Each reading measure in a study was coded as: word reading (cor-
study [1] or N of 1 study [0]), teacher effects controlled (teacher rect reading of words or nonsense words), reading fluency (rate of
effects controlled through random assignment of instructors to con- reading connected text), reading vocabulary (e.g., knowledge of
ditions or each instructor teaching all conditions [1] or teacher the meaning of words in text), reading comprehension (e.g., retells,
effects not controlled [0]), measurement reliability established (reli- cloze tests, questions about text read), reading achievement (norm-
ability was .70 or greater [1] or reliability was less than .70 [0]), referenced measures with total score for reading), and reading
norm-referenced test (norm-referenced measure [1] or not a norm- motivation (e.g., reading efficacy, attitudes, or other motivational
referenced measure [0]), and attrition less than 20% (attrition less beliefs).
than 20% [1] or attrition 20% or greater [0]). In addition to describing and coding writing and reading mea-
Each study was further scored as categorical variables for the fol- sures, a description of the writing treatments was generated. To cat-
lowing four study features which acted as moderators: student type egorize writing treatments, we drew upon instructional
(a score of 1 for typical students in a class; a score of 0 for students categorizations used in previous meta-analyses (Graham et al.,
who were less capable writers [e.g., the 25th percentile or below and 2015; Graham & Perin, 2007; Hillocks, 1986; Koster et al., 2015).
 1012 GRAHAM ET AL.

Figure 1
Preferred Reporting Items for Systematic Reviews and Meta-Analyses Chart
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Table 1 presents and defines these writing treatments and provides a Calculation of ESs
description of which aspect of the WWC model (Graham, 2018) it
was likely to enhance. We did not include in Table 1 other writing ESs for each writing and reading outcome in a study were calcu-
treatments that were tested so infrequently that we were unable to lated as the difference between the average performance of treatment
compute an average-weighted ES. This included treatments specifi- and control students at posttest divided by the pooled SD for the two
cally designed to teach capitalization/punctuation skills, planning groups. Before calculating an ES, posttest scores for each condition
instruction (not already classified as strategy instruction or prewrit- were adjusted by subtracting the average pretest score from the respec-
ing activities), curriculum-based assessment programs, and the tive posttest score. Identical procedures were used to calculate main-
effects of individualized instruction. tenance writing and reading outcomes. The only exceptions to the
Finally, for all writing and reading outcomes, data needed to cal- pretest adjustment involved experimental studies with randomization
culate an ES was recorded (Ms, SD, and sample sizes). If needed data but no pretest measures. This occurred in 76 studies (1,045 ESs).
was missing, other data for calculating ESs was recorded (e.g., F and Furthermore, it was necessary to compute ESs from 29 studies (e.g.,
MSE statistics for analysis of variance). Burks, 2010) by estimating missing SD or Ms. All ESs were computed
Each study was dual coded for the information above by two as Hedge’s g. In some instances, it was necessary to reverse the direc-
trained members of the research team. The only exception involved tion of the ES (e.g., the outcome assessed writing errors).
identifying the type of writing or reading measures (e.g., writing Calculated ESs were examined to identify outliers. An outlier was
quality or reading comprehension, respectively) and the categoriza- defined using Tukey’s (1977) definition: any ES falling 1.5 inter-
tion of writing treatments. This was done for all studies by the first quartile ranges (75th to 25th percentile) above or below the 25th
author who has over 40 years of experience as a writing researcher and 75th percentile. Using this definition, it was necessary to winso-
and another member of the research team. The mean agreement rize three ESs of −7.26, −4.76, and −4.68 in Head (2000), two ESs
between coders was high (94% agreement). Differences were of 18.70 and 11.42 in Logsdon and Leggitt (1939), and one ES in
resolved through discussion. Graner (2007) of 32.14 so they did not exceed Tukey’s criterion.
 SECONDARY WRITING INSTRUCTION 1013

Statistical Analyses The third metaregression included a single variable: time teaching
writing. We would have preferred to include this covariate in the sec-
Our analyses relied on RVE. A random effects model was speci- ond moderator analysis, but almost two in every five studies did not
fied, and all calculations and statistical analyses involving average- provide the information needed to calculate this variable. This would
weighted ESs as well as metaregression moderator analyses were have compromised the second metaregression, as it would have sig-
conducted in the RStudio open-source software (Version 3.6.3; R nificantly limited the number of independent comparisons and ESs
Core Team, 2018; Version 1.2.5033; RStudio Team, 2016) using involved in the analysis. The covariate, time teaching writing, was
functions available in the Robumeta package (Version 2.0; Fisher centered at .17 hr (the smallest time spent on teaching writing).
et al., 2017). Given that many studies included in this review Research Question 3. An average-weighted ES was computed
reported multiple ESs (derived from multiple measures) for the for all 23 writing treatments defined in Table 1. This included
same subsample and had great variability in their sample sizes, we average-weighted ESs for all writing measures combined and each
employed RVE to handle our correlated and dependent data structure writing outcome separately when the degrees of freedom for a mea-
with small sample correction. It should be noted that RVE does not sure was five or greater. Each of these treatments had been tested in at
model heterogeneity at multiple levels nor does it provide corre- least four studies. For five writing treatments, separate average-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

sponding hypothesis tests. Moreover, the power of the categorical

This document is copyrighted by the American Psychological Association or one of its allied publishers.

weighted ESs were presented for different forms of each treatment.

moderator highly depends on the number of studies and features This included comprehensive writing programs (process writing
of the covariate (Tanner-Smith et al., 2016). When the number of approaches associated with the NWP, process writing approaches
studies is small, the test statistics and confidence intervals based not associated with the NWP, and comprehensive writing programs
on RVE can have inflated Type I error (Hedges et al., 2010; that were not process approaches), digital writing (word processing
Tipton & Pustejovsky, 2015). Consequently, some of our modera- and word processing plus programs), strategy instruction (SRSD and
tion analysis results should be interpreted with caution as RVE non-SRSD strategy instruction), transcription instruction (spelling
requires a sufficiently large sample size and may yield inaccurate and handwriting), and feedback (feedback from teacher, peers, com-
p values when degrees of freedom are under 4. puter, and self).
For all average-weighted ESs reported, a confidence interval was Research Question 4. For the 23 specific writing treatments, a
calculated. We further calculated I 2 to describe the ratio of true het- metaregression was conducted to determine if the 10 writing out-
erogeneity to total variance. An I 2 of 0%–40% might not be impor- comes described earlier were differentially impacted. Each analysis
tant, 30%–60% may represent moderate heterogeneity, 50%–90% was contrast coded to compare writing quality to the other writing
indicates substantial heterogeneity, and 75%–100% considerable outcomes. To be included in this analysis, writing quality and at
heterogeneity (Higgins & Green, 2011). least one other writing outcome had to be tested four or more
times. No moderator analysis was conducted for transcription
Analysis for Specific Research Questions
instruction, vocabulary instruction, imagery, inquiry, motivation,
Research Question 1. An average-weighted ES was computed observation, summary writing instruction, creativity/critical think-
for all writing outcomes across all studies. Average-weighted ESs ing, procedural facilitation, sentence instruction, nor revision
were also computed for the 10 writing outcomes (see Table 2). because this criterion was not met.
Research Question 2. Metaregression applying RVE with ran- Additional moderator analyses were conducted for specific writ-
dom effects was used to examine potential moderating effects. Three ing treatments to determine if magnitude of effects were related to
separate metaregressions were conducted for writing outcomes. The grade level (centered on Grade 6) and assessment timing (posttest
first metaregression involved eight indicators of study quality which vs. maintenance). These analyses were not conducted for imagery,
were entered into the model together. These predictors were dummy inquiry, motivation, creativity/critical thinking, or vocabulary
coded as follows: research design (experimental compared to instruction as there were less than four independent comparisons.
quasi-experiment), publication type ( journal article compared to Additionally, assessment timing was not included in many of the
nonjournal article), norm-referenced test (yes or no), treatment fidel- metaregressions because maintenance was assessed less than four
ity (treatment fidelity established compared to not established), N of times in many of the writing treatments.
1 study (not N of 1 study compared to N of 1 study), teacher effects Research Question 5. An average-weighted ES was computed
controlled (teacher effects controlled compared to teacher effects not for all reading outcomes across all studies. Average-weighted ESs
controlled), attrition rate (attrition rate greater than 20% compared to were also computed for three specific reading outcomes: reading
less than 20%), and measurement reliability (measurement reliability achievement, reading comprehension, and reading vocabulary (see
.70 or higher compared less than .70). Table 2).
The second metaregression included five study characteristics Research Question 6. The same three metaregressions applied
entered into the model together. These were grade level (centered with writing outcomes (Research Question 2) were conducted for
at Grade 6), type of student (typical students [full range as well as studies with one or more reading measures. The only exceptions
average and high achieving students] compared to less capable stu- involved the first metaregression, which did not include the covariate
dents [poor writers and students with special needs]), writing during for attrition greater than 20%, and the second metaregression which
treatment (students did not compose during the writing treatment did not include a covariate for assessment timing. No studies with
compared to composing during the writing treatment), assessment reading outcomes reported attrition greater than 20%, and just two
timing (posttest compared to maintenance), and control condition included maintenance outcomes for reading. For the third metare-
type (contrast coded to compare BAU to alternative treatment, gression, time teaching writing was centered at .33 hr (the smallest
diluted treatment, no treatment, writing alternative treatment, coun- time spent on teaching writing when a reading measure was included
ter treatment, and reading treatment). in a study).
 1014 GRAHAM ET AL.

Table 2
Average-Weighted Effect Sizes Overall, Specific Measures, and Specific Writing Treatments
Outcome and treatment N/n K

g (SE) 95% CI p I2
By writing and reading
All writing outcomes 406/352 3,321 0.49*** (.03) [.43, .55] ,.001 86.53
All reading outcomes 52/43 193 0.22** (.07) [.08, .37] .003 82.67
By specific measures
Writing quality 270/231 625 0.46*** (.04) [.39, .53] ,.001 83.28
Ideation 138/128 430 0.46*** (.06) [.35, .57] ,.001 86.11
Elements 60/55 447 0.60*** (.09) [.43, .77] ,.001 89.32
Organization 84/77 265 0.51*** (.09) [.34, .68] ,.001 86.98
Voice 35/31 80 0.30** (.09) [.11, .48] .003 77.01
Process 32/31 321 0.82*** (.13) [.55, 1.09] ,.001 87.4
Translation 113/99 588 0.51*** (.07) [.37, .64] ,.001 86.01
Transcription 44/40 179 0.49** (.14) [.21, .78] .001 90.45
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Knowledge 14/14 104 0.94** (.29) [.32, 1.57] .006 92.97

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Motivational beliefs 66/56 291 0.13** (.04) [.05, .20] .001 49.33
Reading achievement 8/7 25 −0.25 (.34) [−1.06, .56] .50 89.69
Reading comprehension 45/37 143 0.26*** (.07) [.11, .41] ,.001 79.55
Reading vocabulary 10/8 18 0.22 (.24) [−.32, .75] .38 90.13
By writing treatment
Comprehensive programs 41/35 181 0.47*** (.10) [.26, .68] ,.001 87.65
Comprehensive 7/7 44 0.31 (.16) [−.08, .70] .10 80.93
Process not NWP 24/20 94 0.65** (.18) [.28, 1.02] .002 91.70
Process NWP 10/8 43 0.24* (.06) [.06, .42] .02 31.32
Strategy instruction 67/58 609 0.76*** (.09) [.58, .94] ,.001 89.91
SRSD 26/25 261 0.94*** (.19) [.56, 1.33] ,.001 92.06
Not SRSD 42/34 348 0.66*** (.09) [.47, .85] ,.001 87.95
Digital writing 40/35 258 0.31*** (.08) [.14, .48] ,.001 79.85
Word processing 25/22 181 0.22* (.09) [.03, .42] .03 71.51
Word processing plus 14/13 73 0.42* (.17) [.05, .80] .03 87.35
Transcription instruction 32/28 140 0.71** (.23) [.25, 1.17] .004 93.64
Spelling 28/24 115 0.53* (.20) [.12, .93] .01 92.13
Handwritinga 3/3 20 3.31 (1.51) [−3.17, 9.79] .16 98.02
CAI 15/13 66 0.32** (.10) [.10, .53] .007 70.71
Creativity/critical thinking 10/6 24 0.27* (.08) [.08, .46] .012 39.53
Emulation 12/12 82 0.46** (.13) [.17, .76] ,.001 71.66
Feedback 41/36 472 0.34*** (.09) [.16, .51] ,.001 85.12
From teacher 16/13 187 0.41** (.12) [.15, .68] .005 79.10
From peer 14/13 119 0.44* (.18) [.04, .83] .03 90.39
From computer/machine 8/7 70 0.13 (.13) [−.19, .44] .38 81.47
From self 6/6 90 0.08 (.12) [−.23, .38] .54 57.29
Setting goals 13/11 178 0.44** (.11) [.20, .69] ,.01 71.60
Prewriting 30/28 246 0.49*** (.08) [.32, .67] ,.001 83.64
Procedural facilitation 12/9 59 0.24 (.15) [−.08, .56] .12 82.57
Grammar instruction 13/12 82 0.77** (.19) [.36, 1.17] .002 89.55
Sentence instruction 14/13 112 0.73*** (.17) [.37, 1.09] ,.001 84.05
Imagery 5/5 15 0.44 (.16) [−.01, .89] .05 69.99
Inquiry 4/4 7 0.92* (.19) [.28, 1.57] .02 74.37
Enhancing motivation 8/8 28 0.18 (.19) [−.27, .63] .37 81.21
Observation 7/7 121 0.41* (.12) [.11, .71] .02 74.79
Peer assistance 27/23 170 0.51*** (.09) [.33, .69] ,.001 88.71
Revision instruction 8/8 82 0.24 (.19) [−.20, .69] .23 86.48
Summarization instruction 11/10 104 0.49* (.20) [.05, .93] .03 78.83
Text structure instruction 13/12 95 0.39* (.16) [.05, .73] .03 89.78
Vocabulary instruction 5/4 67 0.34 (.35) [−.63, 1.31] .38 90.67
Increased writing 20/13 65 0.14 (.10) [−0.07, .35] .18 63.4
Note. N refers to the counts of independent samples nested within n, which represents the counts of studies; K refers to
the number of effect sizes; CI = confidence interval; NWP = National Writing Project; SRSD = self-regulated strategy
development; CAI = computer-assisted instruction.
a
df , 4; results should be taken with caution.
* p , .05. ** p , .01. *** p , .001.

Publication Bias Transparency and Openness Statement

We tested for possible publication bias by examining graphical We reported the exclusion and inclusion criteria for selection of
representations of the funnel plot for all outcomes in all studies. studies included in the review as well as the search procedures
An Egger’s test was further conducted with the observed ESs to implemented. All coding as well as data-analytic procedures
determine the possibility of publication bias. were reported, and the coded data is available from the first
 SECONDARY WRITING INSTRUCTION 1015

author. This meta-analysis and the subsequent analyses were not (2%), vocabulary instruction (2%), increased writing (2%),
preregistered. computer-assisted instruction (2%), other (2%), and procedural
facilitation (2%). All other writing treatments each accounted for
Results less than 1% of all interventions. The most common control condi-
tion was BAU (37%), with other control conditions occurring less
Characteristics of Studies
often: counter treatment (18%), diluted writing treatment (17%),
The earliest study from the 357 documents was a thesis con- writing alternative treatment (10%), no treatment (9%), reading
ducted in 1932 by Hamilton examining the effects of spelling treatment (7%), and alternative treatment (3%).
instruction. From 1932 to the end of 1950, we located only six stud- Sixty-three percent of the documents provided enough informa-
ies that met our inclusion criteria. During the 1960s and 1970s, tion to determine the amount of time spent teaching writing. For doc-
there was a sizeable increase in writing intervention studies with uments that provided this information, an average of 13.56 hr was
secondary students (14 and 20 documents, respectively), but such devoted to instruction (SD = 15.99 hr).
research became increasingly common in subsequent decades: 61 It was possible in 99% of documents to determine the content area
documents in the 1980s, 51 in the 1990s, 73 in the 2000s, and in which a targeted writing treatment was tested. This most com-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

114 in the 2010s. monly occurred in language arts (87%), followed by social studies
Of the 349 documents that reported location, 69% (n = 242) took (7%), science (3%), math (1%), and other content areas (2%).
place in the United States. The most common locations outside of the Likewise, it was possible to determine who led instruction in 98%
United States were in order: Europe (n = 55), Asia (n = 21), Middle of documents. This most commonly involved teachers (43%), fol-
East (n = 15), and Canada (n = 14). In 62 of the documents (18%), lowed by research staff (22%); teachers and technology (9%); teach-
the teaching of writing involved a language other than English. ers and research staff (9%); peers (7%); research staff and technology
Writing in 50 of these 62 documents involved an alphabetic lan- (4%); technology (3%); and teacher, research staff, and technology
guage (12 involved writing in Chinese). (4%).
Although the type of community where studies were conducted It was further possible in 95% of documents to determine if stu-
was not reported in 121 documents (34%), when this was reported dents engaged in writing during the writing treatment and, if they
urban communities were most common (45%), followed by subur- did so, in what genre. For documents that provided this information,
ban (34%), rural (10%), and some combination of communities students in the treatment wrote as part of instruction in 84% of the
(10%). Researchers only indicated SES in 16% of documents. In documents. The most common single forms of writing they were
these cases, middle and low SES communities were most common asked to undertake were persuasive (38%) and expository (16%),
(42% each), with the remaining studies mostly occurring in a mix followed by narrative writing (7%) and the singular focus on other
of SES communities (14%). forms of writing such as summary, journal entries, and poems
The studies contained in the 357 documents involved 52,984 stu- (9%). Close to one-fourth (23%) of all documents when students
dents in Grades 6–12. Grade level (or age) of students was provided wrote during the writing treatment condition involved multiple
in all documents. Forty-seven percent of studies involved middle genres.
school students, 45% high school students, and 8% both. Most stud- Professional development or training was provided to people
ies were conducted with students in a single grade (64%). Thirteen delivering the writing treatment in two-thirds of the documents.
percent of studies involved two grades, 10% involved three grades, For documents that provided this information, professional develop-
and 13% involved four or more grades. Most documents (94%) also ment/training was provided 55% of the time; it was not provided 2%
provided enough information to classify participating students. of the time; it was provided for some but not all instructional agents
When this was the case, student samples mostly reflected the full 2% of the time; and it was not relevant 41% of the time (technology
range of typical classrooms (70%), with the remaining samples provided instruction or the study included a single researcher who
split between average writers (5%), high achieving writers (3%), served as the teacher).
less capable writers (6%), students with special needs (7%), and Studies in all 357 documents included at least one writing
some combination of these categories (9%). measure, but just 12% of them included at least one reading
Student gender, race, and SES were not reported in 41%, 53%, measure. The most common measures were writing quality (17%),
and 90% of documents, respectively. When gender was reported, genre elements (13%), ideation (8%), organization (8%), syntax
51% of students were girls. When race was reported, 50% of students (7%), revising (5%), writing output (4%), spelling (4%), reading
were White, 24% Hispanic, 11% Asian, 11% Black, 1% First comprehension (4%), vocabulary (4%), attitudes toward writing
People, and 2% other (e.g., mixed-race). Only 24% of the documents (3%), conventions of writing (3%), efficacy for writing (3%), knowl-
indicated that participating students were L2 or both L2 and L1. For edge of writing (3%), sentence fluency (3%), voice (2%), writing
documents that provided this information, 46% of the documents motivation (2%), self-regulation (2%), planning writing (1%), and
focused just on L2 students. writing perseverance (1%). All other measures each accounted for
The writing treatments tested in the obtained documents were var- less than 1% of all assessments. Most writing outcomes were admin-
ied. Across all writing interventions, the most common treatments istered immediately following the end of the writing treatment
were strategy instruction (18%), feedback (13%), digital writing (81%). Maintenance was assessed in 19% of all studies. Writing
(7%), prewriting activities (7%), process writing approach (5%), maintenance assessments were mostly limited to 10 writing treat-
goal setting (5%), peer assistance and support (5%), spelling instruc- ments: strategy instruction (16% of all maintenance outcomes), dig-
tion (4%), observation (4%), summarization instruction (4%), sen- ital writing (14%), feedback (9%), goal setting (8%), text structure
tence instruction (3%), text structure instruction (3%), emulation instruction (7%), teaching transcription skills (6%), computer-
of models (3%), grammar instruction (2%), revision instruction assisted instruction (6%), comprehensive writing programs (6%),
 1016 GRAHAM ET AL.

summarization instruction (5%), and increasing writing (4%). conditions and both were assumed to be effective, we did not include
Reading maintenance was assessed in only two studies. them in the analyses.
The 352 documents included in our analyses produced 114 unique
Quality of Studies treatment/control comparisons among 406 independent samples,
with a total of 3,514 ESs (3,321 of these ESs involved writing out-
Forty-five percent of studies were published in a peer-reviewed
comes and 193 reading outcomes) involving 52,529 students.
journal. Of the remaining studies, 48% were dissertations and thesis,
Table 1 in the online supplemental materials includes the following
3% research reports and ERIC documents, 2% book chapters, 1%
information for each independent comparison in the 352 documents:
unpublished studies from authors, and 1% conference papers.
reference, grade level of students, type of student, writing treatment,
Fifty-six percent of studies were experiments (31% true experiments
control condition, and aggregated ES and variance for writing as
and 25% randomized control designs). Forty-four percent of studies
well as reading when available. Table 2 in the online supplemental
were quasi-experiments.
materials includes the five studies excluded for our analysis (see
Sixty percent of studies were classified as a not N of 1 study. In
above). These tables as well as references for all studies in the meta-
other words, both treatment and control conditions were composed
analysis (including studies referenced in the Results section) are
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

of more than one group of students. It was not possible to determine

This document is copyrighted by the American Psychological Association or one of its allied publishers.

included in the online supplemental materials.

N of 1 status in 3% of investigations.
Teacher effects were controlled in 56% of studies. In 4% of stud- Research Question 1: Does Teaching Writing Improve
ies, teacher effects were not relevant because instruction was deliv- Writing Outcomes?
ered via computer or by peers. Although teacher effects were clearly
not controlled in 30% of studies, there was not enough information As can be seen in Table 2, when all writing outcomes were con-
provided to determine if teacher effects were controlled in 10% of sidered collectively, teaching writing resulted in a statistically signif-
investigations. icant improvement in students’ performance (ES = 0.49; 74.41% of
Attrition greater than 20% was not a problem in 97% of studies. ESs were positive). Furthermore, teaching writing resulted in statisti-
Similarly, reliability was established for 97% of measures across cally detectable improvements for all 10 writing outcomes exam-
studies. In contrast, treatment fidelity was only established in 31% ined. The most common writing outcome, writing quality,
of studies. When it was established, this most commonly occurred improved by 0.46 SDs, as did the writing outcome ideation.
through observation (41%), followed by a combination of observa- Teaching writing improved five specific writing outcomes by more
tion, teacher, and researcher support (35%). In 13% and 11% of than one-half a SD: knowledge of writing (0.94), writing processes
studies when fidelity was established, this occurred through (0.82), genre elements (0.60), organization (0.51), and translation
researcher claims and teacher reports, respectively. (0.51). The average-weighted ESs for the other three writing out-
comes, transcription, voice, and motivation, were 0.49, 0.30, and
Included and Excluded Writing Comparisons 0.13, respectively.
Although all of the studies included in the 357 documents Research Question 2: Do Study Attributes Predict
involved experiments or quasi-experiments (with pretests) examin-
Magnitude of Writing Outcomes?
ing the impact of one or more writing treatments, we excluded five
studies from the analyses because it was not clear which writing For all writing outcomes collectively, three moderator analyses were
interventions were treatment and control in these studies. The five conducted to determine if study attributes were related to variability in
excluded studies were Hamilton (1932) who compared the effective- effects. There was considerable variability in ESs, as 86.53% of the
ness of two commonly used preexisting spelling programs, as well as variance was due to between-study factors (see Table 2).
Sineath (2014) who compared a prewriting activity using one’s own The first moderator analysis examined if measures of study quality
social media data versus a prewriting activity using a friend’s social indicators (e.g., research design) were related to the magnitude of
media data. It also included Dunnagan (1990) who compared a pre- writing effects (Table 3). Three study quality indicators, publication
writing activity involving role playing to students emulating a model type, norm-referenced test, and not N of 1 study, were statistical con-
of good writing; Stark and Nagy (2011) who compared spelling and tributors to the model with α set at .05 and variance due to the other
grammar instruction; and Kieft et al. (2008) who compared revising study quality indicators controlled. Journal articles were associated
the plan for a first draft of a paper to revising the first draft of that with larger effects than nonjournal articles ( journal articles were
paper. .21 ES larger), but studies that employed norm-referenced tests or
Some comparisons in other studies included in this meta-analysis were not N of 1 were associated with smaller effects (not norm-
were not included for similar reasons. For example, Zimmerman and referenced comparison was .15 ES lower and not N of 1 comparison
Kistantas (1999) compared multiple process and product goals for was .20 ES lower).
revising. It was possible to isolate the effects of each of these The second moderator analysis examined if specific study charac-
goals (i.e., outcome, process, and shifting goals) because they teristics (e.g., grade level) were related to magnitude of writing
could be compared to a no goal control condition, but we did not effects. These study characteristics were not statistically associated
compare each goal (e.g., outcome) to the other (e.g., process) with variation in writing effects when alpha was set at .05 (see
because they were each likely to be effective and none of them Table 3).
were considered a control condition. Similarly, Hare and The third moderator, time teaching writing, did not make a statisti-
Borchardt (1984) compared two forms of summarization instruction cally significant contribution to predicting variance in writing ESs.
(inductive and deductive) to a no-treatment control. Again, because This analysis included 250 independent comparisons with 2230
these two forms of summarization instruction were not control ESs (average-weighted ES = 0.00).
 SECONDARY WRITING INSTRUCTION 1017

Table 3
Moderation Analyses for Study Quality Indicators and Study Characteristics for All Measures
Outcomes N/n K Estimate (SE) I2
Writing outcomes
Full model—study quality indicators 401/349 3,297 85.48
Intercept .50* (.20)
Experimental design −.01 (.07)
Journal .21*** (.06)
Norm-referenced test −.15* (.07)
Treatment fidelity established .06 (.06)
Not N of 1 study −.20** (.06)
Teacher effects controlled .01 (.06)
Attrition less than 20% .04 (.09)
Measure reliable −.04 (.19)
Full model—study characteristics 331 2,856 85.28
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Intercept .94*** (.20)

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Grade-level −.03 (.02)

Less capable writers .06 (.11)
Writing during treatment −.23 (.14)
Maintenance .03 (.09)
Control alternative .06 (.23)
Control diluted −.05 (.11)
Control no treatment .03 (.14)
Control other −.15 (.11)
Control counter −.16 (.08)
Control reading −.13 (.15)
Reading outcomes
Full model—study quality indicators 51/42 190 83.27
Intercept .42 (.30)
Experimental design .25 (.22)
Journal −.02 (.17)
Norm-referenced test −.24 (.18)
Treatment fidelity established −.11 (.20)
Not N of 1 study −.01 (.17)
Teacher effects controlled .09 (.17)
Measure reliable −.26 (.27)
Full model—study characteristics 48 170 81.91
Intercept .15 (.45)
Grade-level .07 (.04)
Less capable writers −.21 (.15)
Writing during treatment −.47 (.25)
Control alternative .07 (.22)
Control diluted −.28 (.24)
Control no treatment −.07 (.16)
Control counter .20 (.24)
Control reading −.21 (.17)
Control other .17 (.14)
Note. N is the count of independent samples nested within n, which represents the counts of studies;
K refers to the number of effect sizes.
* p , .05. ** p , .01. *** p , .001.

Research Question 3: Do Specific Writing Treatments Comprehensive Writing Programs. Comprehensive writing
Improve Writing Outcomes? programs had a statistically detectable impact on students’ writing
when all writing outcomes were aggregated (ES = 0.47), resulting
Average-Weighted ES for 23 Specific Writing Treatments in a positive outcome for 80% of the ESs. Similar effects were
The outcomes for the 23 different writing treatments are pre- obtained for writing quality. The average-weighted ES was 0.49
sented in Table 2. The average-weighted ES for each writing treat- (see Table 4), with positive outcomes obtained for 86% of ESs.
ment in Table 2 was based on all writing outcomes testing that There was considerable variability in ESs (88% and 85%,
treatment. Table 4 presents outcomes for writing quality for respectively).
each writing treatment. Average-weighted ESs for writing quality Twenty-four of the comprehensive writing program comparisons
were not calculated for the following treatments because this tested the effectiveness of the process writing approach. This
outcome was not assessed frequently enough: transcription approach resulted in statistically detectable improvements in stu-
instruction, imagery, inquiry, enhancing motivation, observation, dents’ writing on all measures (0.65; Table 2 [84% of ESs were pos-
revising instruction, and summary instruction. itive]) and writing quality specifically (0.61; Table 4 [92% of ESs
 1018 GRAHAM ET AL.

Table 4
Writing Treatment Average-Weighted Effect Sizes for Writing Quality
Writing treatment N/n K

g (SE) 95% CI p I2
Comprehensive programs 36/30 51 0.49*** (.10) [.28, .70] ,.001 84.79
Comprehensive 6/6 12 0.58 (.32) [−.23, 1.39] .13 89.74
Process not NWP 19/15 25 0.61** (.16) [.26, .95] .002 88.32
Process NWPa 10/8 13 0.23* (.07) [.03, .44] .03 31.38
Strategy instruction 58/50 135 0.93*** (.12) [.69, 1.17] ,.001 90.22
SRSD 22/21 58 1.16*** (.24) [.67, 1.64] ,.001 91.17
Not SRSD 37/30 77 0.79*** (.13) [.53, 1.05] ,.001 89.12
Digital writing 34/30 63 0.21* (.08) [.05, .38] .01 66.95
Word processing 22/20 37 0.15 (.11) [−.07, .38] .16 63.64
Word processing plus 12/10 24 0.26 (.15) [−.06, .58] .10 74.47
CAI 10/9 19 0.30 (.13) [−.00, .60] .05 79.72
Creativity/critical thinking 10/6 12 0.26* (.09) [.05, .46] .02 39.91
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Emulation 8/8 13 0.56* (.15) [.18, .94] .01 49.27

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Feedback 23/21 58 0.27* (.11) [.05, .49] .02 82.57

From teachera 3/3 5 0.28 (.13) [−.28, .84] .16 70.43
From peer 10/9 35 0.53* (.20) [.02, .09] .02 84.52
From computer/machine 6/5 9 0.16 (.19) [−.34, .65] .45 88.90
From selfa 4/4 9 −0.17 (.18) [−.75, .41] .42 60.40
Setting goals 9/7 24 0.49** (.13) [.18, .80] ,.01 38.47
Prewriting 16/15 29 0.44** (.11) [.21, .66] ,.01 72.54
Procedural facilitation 6/5 11 0.58 (.26) [−.09, 1.26] .08 83.98
Grammar instruction 6/5 13 0.47* (.16) [.04, .89] .04 80.81
Sentence instruction 9/9 15 0.75** (.17) [.35, 1.15] ,.01 71.96
Peer assistance 19/16 54 0.43*** (.09) [.23, .62] ,.001 78.01
Revision instruction 7/7 15 0.20 (.20) [−.25, .64] .34 89.49
Text structure instruction 10/9 24 0.39* (.16) [.05, .73] .03 89.78
Increased writing 14/7 21 0.05 (.16) [−.30, .40] .76 71.14
Note. Treatments involving transcription, imagery, inquiry, motivation, observation, summarization, and vocabulary
were omitted due to insufficient power; K refers to the number of effect sizes. CI = confidence interval; NWP =
National Writing Project; SRSD = self-regulated strategy development; CAI = computer-assisted instruction.
a
df , 4; results should be taken with caution.
* p , .05. ** p , .01. *** p , .001.

were positive]), although there was considerable variability in ESs Netherlands, Belgium, and Portugal). All studies applied a
(91.70% for all measures and 88.32% for writing quality). These quasi-experimental design and BAU was the most common control
studies involved students in Grades 6–12 mostly in the United condition.
States (two studies involved less capable writers and four took Strategy Instruction. Strategy instructional studies focused on
place in Canada [2], Nigeria, and Vietnam). Nine comparisons teaching students to plan text (Yeh, 1998), edit and/or revise it
used an experimental design, and control conditions were mostly (Reynolds, 1985), or both (De La Paz & Graham, 2002). Teaching
BAU. students such strategies had a statistically detectable impact on all
Another 10 comparisons tested the effectiveness of process writ- writing outcomes collectively (ES = 0.76; 86% of all ESs were pos-
ing approaches affiliated with the NWP (Table 2), resulting in statis- itive) and with writing quality (ES = 0.93; 96% of ESs were posi-
tically detectable effects of 0.24 for all writing measures (77% of tive). Variability of ESs was large (89.91% and 90.22%,
ESs were positive), and 0.23 for writing quality (85% of ESs were respectively).
positive). Variability in ESs was minimal (31.32% for all measures For all comparisons involving just SRSD strategy instruction, a
and 31.38% for writing quality). These studies involved typically statistically detectable ESs of 0.94 for all writing measures
developing writers in Grades 6–12 in the United States. Two of (Table 2; 89% of ESs were positive) and 1.16 for writing quality
the comparisons applied an experimental design, and the control (Table 4; 95% of all ESs were positive) were obtained. Variability
conditions were always BAU. in ESs was substantial (92.06% and 91.17%, respectively). The
The other seven comparisons, tested commercial writing pro- majority of studies involved typical writers in Grades 6–12, but
grams (e.g., Airhart, 2005), programs teaching students to write less capable writers and students with special needs were a focus
for multiple purposes (e.g., Berge et al., 2019), and programs that of instruction in 10 comparisons. Close to one-half of the compari-
combined writing and skills instruction (e.g., Weinke, 1981). The sons took place outside the United States (Germany, Portugal, Spain,
obtained effects for these seven comparisons were not statistically Canada, Indonesia, Scotland, Italy, Lebanon, and Egypt), and 17 of
greater than no effect for all writing measures (see Table 2) or for them applied an experimental design. The most common control
writing quality (see Table 4). These studies included typical writers condition was BAU.
in Grades 6–12, although two studies included adolescents who were The remaining strategy comparisons (not SRSD) resulted in stat-
less capable writers and another study included students with special istically detectable ESs of 0.66 for all writing outcomes (Table 2;
needs. A majority of these studies were conducted in Europe (the 83% of ESs), and 0.79 for writing quality (Table 4; 96% of ESs
 SECONDARY WRITING INSTRUCTION 1019

were positive). Variability in ESs was large (87.95% and 89.12%, 2013) as well as computer-assisted instruction for basic writing
respectively). These studies mostly involved typical writers in the skills (May, 1985). Such instruction resulted in a statistically detect-
United States in Grades 6–12, but close to one-third of the studies able ES of 0.32 for all writing outcomes (Table 2; 74% of ESs were
took place in Spain, Belgium, Canada, or China, with close to one- positive; variability in ESs was 70.71%), but the ES for writing qual-
fourth of comparisons focused on less capable writers or students ity was not statistically greater than no effect (Table 4). These studies
with special needs. An experimental design was applied in 30 com- were conducted mostly with students in the United States who were
parisons, and the most frequent control conditions were BAU, read- in Grades 6–12 (one study was from Australia and another from
ing, and diluted. Taiwan). Three comparisons involved less capable writers or stu-
Digital Writing. Digital writing studies (word processing, dents with special needs. A third of studies employed an experimen-
word processing plus, and speech-to-text synthesis) resulted in a tal design, and the most common control conditions were counter,
statistically detectable ES of 0.31 for all writing outcomes diluted treatments, and no instruction. Students wrote in all studies.
(Table 2; 66% of ESs were positive) and 0.21 for writing quality Creative and Critical Thinking. An example of creative think-
(67% of 63 ESs were positive; Table 4). There was considerable var- ing instruction was teaching students to use metaphors, similes, and
iability in effects (79.85% and 66.95%, respectively). analogy when writing (Teo & Tan, 2003). Critical thinking instruc-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

For the digital writing comparisons involving just word process- tion was typified by Pisano (1980). In this study, students were asked
ing, a statistically detectable ES of 0.22 for all writing measures inquiry questions to help them think more critically about their writ-
(64% of ESs were positive; variability in ESs was 71.51%; ing. Creative and critical thinking instruction resulted in statistically
Table 2) was obtained, but this was not the case for writing quality detectable ESs of 0.27 for all writing outcomes (Table 2; 75% of ESs
(see Table 4). Students in processing studies were mostly typical were positive) and 0.26 for writing quality (Table 4; 83% of ESs
writers in Grades 6–12 in the United States, although nine studies were positive). Variability in ESs were minimal (39.53% and
focused on less capable writers and students with special needs (a 39.91%, respectively). Studies mostly involved students who were
single study occurred in the Netherlands). Sixteen of the compari- typical writers in Grades 6–12 from the United States (one study
sons applied an experimental design, and the most common control was conducted in England and another one in Singapore). Only
conditions were counter (i.e., compared to writing by hand) and two comparisons applied an experimental design and all control con-
BAU. ditions were BAU. Students composed text in all comparisons.
Examples of treatments in the word processing plus program com- Emulating Models. Examples of emulation included studying
parisons included multimedia word processing that allowed writers and emulating a good persuasive essay (Crowhurst, 1990) and the
to compose with text, graphics, sound, and videos (Carlin-Menter, process of giving good peer feedback (Hübner et al., 2010). Such
2006) and word processing programs with added software such as instruction resulted in statistically detectable ESs of 0.46 for all writ-
a tool that assisted with homophones in writing (Lange et al., ing outcomes (Table 2; 68% of all ESs were positive) and 0.56 for
2009). These studies resulted in a statistically detectable ES of writing quality (Table 4; 92% of ESs were positive). Variability in
0.42 for all writing measures (Table 2; 68% of ESs were positive; ESs were 71.66% and 49.27%, respectively. All but one of the com-
variability in ESs was 87.35%), but the ES for writing quality was parisons involved typical writers in Grades 6–12, with one-half of
not statistically greater than no effect (Table 4). Students in the them occurring in the United States and other one-half in Canada,
word processing plus studies were mostly typical writers in Grades the Netherlands, Germany, and China. Two-thirds of the compari-
6–12 students in the United States, but two studies focused on less sons applied an experimental design, with the most common con-
capable writers and five publications were from the following coun- trols including BAU, no treatment, reading, and diluted treatment.
tries: the Netherlands, Israel, South Korea, Italy, and England. Eight Students composed text in all comparisons.
of the comparisons applied an experimental design, and the most Feedback. Feedback produced statistically detectable ESs of
common control conditions were diluted and counter treatments as 0.34 for all writing measures (Table 2; 65% of all ESs were positive)
well as BAU. and 0.27 for writing quality (Table 4; 67% of 58 ESs positive).
Teaching Transcription Skills. Studies assessing the effec- Variability in ESs were 85.12% and 82.57%, respectively. These
tiveness of teaching transcription skills mostly involved teaching studies mostly involved typical writers in Grades 6–12 in the
spelling to students in Grades 6–12, but three comparisons United States, although studies were conducted in Belgium, the
focused on teaching handwriting in Grades 6–9 and another Netherlands, Spain, Taiwan, and Latavia. Twenty-seven compari-
study tested keyboarding with students in Grades 8 and sons used an experimental design, and control conditions were
9. Transcription instruction (140 ESs) yielded statistically detect- mostly alternative and counter treatments.
able ESs of 0.71 for all writing measures (Table 2). Most ESs were When teacher feedback specifically was examined, a statistically
positive (76%), and variability was 93.64%. Most of these studies detectable ES of 0.41 was obtained for all writing outcomes
took place in the United States, but two occurred in Australia and (Table 2; 71% of ESs were positive; variability in ESs was
one in Greece. Students were mostly typical writers, but six com- 79.10%; Table 2), but not for writing quality (see Table 4). Peer
parisons involved less capable writers or students with special feedback resulted in a statistically detectable ESs of 0.44 for all writ-
needs. Fourteen comparisons applied an experimental design, ing outcomes (60% of ESs were positive) and 0.53 for writing qual-
and the most common control conditions were BAU, reading, ity (69% of ESs were positive; Table 4). Variability of ESs for peer
and counter treatments. Students wrote text in only four of the feedback were 90.39% and 84.52%, respectively. For all writing out-
comparisons. comes and writing quality, statistically detectable ESs were not
Computer-Assisted Instruction. Examples of computer-assisted obtained for self-feedback or computer feedback.
instruction include an intelligent computer system that offered Goal Setting. Examples of goal setting included an objective to
students help with planning, drafting, and revising (Crossley et al., refute a specific number of counterarguments when writing a
 1020 GRAHAM ET AL.

persuasive text (Page-Voth & Graham, 1999) and an objective to (Table 2; 86% of the ESs were positive) and 0.75 for writing quality
minimize number of words when combining smaller sentences (Table 4; 87% of ESs were positive). Variability in ESs were 84.05%
into more complex ones (Zimmerman & Kitsantas, 1999). Goal set- and 71.96%, respectively. Studies mostly involved typical writers in
ting resulted in statistically detectable ESs of 0.44 for all writing out- Grades 6–12 from the United States (one study was from Canada and
comes (Table 2; 80% of ESs were positive) and 0.49 for writing one study involved less capable writers). Seven comparisons applied
quality (Table 4; 88% of ESs were positive). Variability in ESs an experimental design, and the most common control conditions
were 71.60% and 38.47%, respectively. All but one study were BAU and diluted treatments. Students did not compose text
(Canada) took place in the United States. Students were mostly typ- in 10 of the comparisons.
ical writers in Grades 6–11, but three comparisons involved less Imagery. Examples of imagery included using it to activate
capable writers or students with special needs. Goal setting was knowledge for writing (Giles, 1989) and the use of imagery to
tested with an experimental design in nine comparisons, and the learn spellings by forming images of words (Weintraub, 1996).
most common control conditions were diluted and BAU. Students Imagery instruction did not result in a statistically detectable ES
composed text in all but one study. for all writing outcomes (see Table 2). Studies mostly involved typ-
Prewriting Activities. Examples of prewriting included brain- ical writers in the United States in Grades 6–12 (one study was con-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

storming ideas and small group discussion (Vinson, 1980), using a ducted in Saudi Arabia). Two comparisons applied an experimental
graphic organizer to generate and organize ideas (Bulgren et al., design, and the most common control condition was diluted treat-
2009), and reading to obtain writing content (van Driel et al., ment. Students composed text in two comparisons.
2022). Such activities resulted in statistically detectable ESs of Inquiry. Examples of inquiry included examining different sce-
0.49 for all writing outcomes (Table 2; 73% of ESs were positive) narios of related concepts to determine how they differ in order to
and 0.44 for writing quality (Table 4; 69% of ESs positive). write about them (Hillocks et al., 1983), and formulating hypotheses
Variability in ESs were 83.64% and 72.54%, respectively. Studies as well as gathering and analyzing data in order to make decisions on
mostly involved typical writers in Grades 6–12, although two com- how to structure written text (Widvey, 1971). Inquiry resulted in a
parisons included less capable writers. The majority of studies were statistically detectable ES of 0.92 for all writing outcomes
conducted in the United States, but comparisons occurred in Canada, (Table 2; all ESs were positive), although there was considerable
China, the Netherlands, Portugal, Germany, Slovenia, Cyprus, and variability in ESs (74.37%). All comparisons involved typical writ-
Taiwan. Most comparisons (20) applied an experimental design, ers in Grades 8–12 in the United States. One-half of the studies
and the most common controls were BAU, counter, and diluted. applied an experimental design, and BAU was the most common
Students composed text in all studies. control condition. Students composed text in all comparisons.
Procedural Facilitation. Examples of procedural facilitation Enhancing Motivation. Examples of procedures to enhance
included a study by Carvalho (2002). In this study, students were motivation included attribution retraining (Fulk, 1996), situating
directed to use cards that contained prompts for evaluating a compo- writing in familiar and real-life contexts (Hwang et al., 2014), and
sition (e.g., I am repeating myself) and making a revision (e.g., I am teaching students to give themselves positive self-evaluations
going to add…). Procedural facilitation did not produce statistically about their writing (Luxin & Rong, 2012). These methods did not
detectable ESs for either all writing outcomes or writing quality (see result in an ES statistically greater than no effect for all writing out-
Tables 2 and 4). Studies mostly focused on typical writers in Grades comes (see Table 2). Studies mostly involved typical students in
6–12 (two studies involved students with special needs), and they Grades 6–12, although three studies focused on students with special
were conducted in multiple countries (United States, China, the needs. The most common location of studies was the United States,
Netherlands, Portugal, Chile). Eight of the comparisons applied an but studies were also conducted in Chile, Lebanon, Taiwan, and
experimental design, and the most common control conditions Hong Kong. Only three of the comparisons applied an experimental
were no treatment, counter treatment, and writing alternative treat- design, the most common control condition was diluted treatments,
ment. Students composed text in all but one comparison. and text was composed in all but two comparisons.
Grammar Instruction. Grammar instruction ranged from Observation. Examples of observation included observing
teaching specific aspects of grammar and how to use this knowledge writers as they composed and observing other students read and
(Myhill et al., 2018) to decontextualized instruction (Erlam, 2003). react to text (Couzijn, 1999). Such observations resulted in a statisti-
Such instruction resulted in statistically detectable ESs of 0.77 for all cally detectable ES of 0.41 for all writing measures (Table 2; 80% of
writing outcomes (Table 2; 93% of ESs were positive) and 0.47 for ESs were positive). Variability in ESs was 74.79%. These studies
writing quality (Table 4; all ESs were positive). Variability in ESs involved typical students in Grade 6–9 from the Netherlands and
were 89.55% and 80.81%, respectively. Studies focused on typical Spain, although two comparisons were conducted in the United
writers in Grades 6–12 (one study involved less capable writers), States. All studies applied an experimental design, and the most
with six documents originating in the United States and the remain- common control conditions were diluted, counter, and writing alter-
der in England, Iran, Belgium, and New Zealand. Six comparisons native. Compositions were written in all comparisons.
applied an experimental design, and the most common control con- Peer Assistance. Peer assistance mostly involved students
ditions were BAU and diluted treatment. Students did not compose working with another student to plan, edit, and/or revise text (e.g.,
text in five of the comparisons. De Smedt et al., 2019; Prata et al., 2019). Peers helping each other
Sentence Instruction. Examples of sentence instruction ranged produced statistically detectable ESs of 0.51 for all writing outcomes
from sentence combining (O’Hare, 1971) to explicitly teaching stu- (Table 2; 82% of ESs were positive) and 0.43 for writing quality
dents how to construct specific sentence structures through explana- (Table 4; 87% of ESs were positive). Variability in ESs was
tion, models, and practice (Kennedy, 2008). Such instruction 88.71% and 78.01%, respectively. Studies mostly included typical
produced statistically detectable ESs of 0.73 for all writing outcomes students in Grades 6–12, although less capable writers were included
 SECONDARY WRITING INSTRUCTION 1021

in three studies. Although a majority of comparisons occurred in the (see Tables 2 and 4). The impact of increased writing was mostly
United States, studies assessing peer assistance were conducted in tested with typical writers in Grades 6–12 in the United States
Belgium, the Netherlands, Iran, Pakistan, Indonesia, and Israel. (one study included less capable writers and one publication was
Only eight of the comparisons applied an experimental design, from South Korea). Ten of the comparisons applied an experimental
and the most common control conditions were BAU and counter design, and the most common control conditions were BAU and no
treatments. Students composed text in all but three of the instruction.
comparisons.
Revision Instruction. Examples of revision instruction Research Question 4: Do Study Attributes Predict
included directly teaching students how to add, delete, substitute, Magnitude of Writing Outcomes for Specific Writing
reword, and move text (Barkel, 1990) and teaching students to revise Treatments?
substantive aspects of text (Head, 2000). Such instruction did not
result in an ES statistically greater than no effect for all writing mea- Only two of the 12 analyses conducted to determine if specific
sures and writing quality (see Tables 2 and 4). Teaching students to writing treatments differentially impacted writing outcomes pro-
revise was assessed in eight comparisons. These studies involved duced a statistically significant outcome (see Table 5). For strategy
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

instruction, two writing outcomes, translation and motivation,

This document is copyrighted by the American Psychological Association or one of its allied publishers.

typical students in Grades 6–10 in the United States, with the excep-
tion of one study conducted in Hong Kong. Three of the studies were statistical contributors to the model. Both of these outcomes
employed an experimental design, and control conditions mostly were associated with smaller average-weighted effects when com-
involved BAU or a diluted writing treatment. Students composed pared to writing quality: The average-weighted ESs for motivation
text in all comparisons. and translation were, respectively, 0.67 and 0.53 SDs smaller than
Summarization Instruction. Examples of summarization the one for writing quality. For digital writing, motivation was a stat-
instruction included teaching a rule-based summarization strategy istical contributor to the model. Motivational outcomes were associ-
(Bean & Steenwyk, 1984) and implicit instruction aimed at helping ated with a smaller average-weighted ES when compared to writing
students acquire the needed information for writing a summary quality (0.31 compared to −0.15). The 10 metaregressions for writ-
(Hare & Borchardt, 1984). Such instruction produced a statistically ing outcomes that did not find statistically detectable differences
detectable ES of 0.49 for all writing measures (Table 2: 70% of ESs between writing quality and other writing outcomes are presented
were positive). Variability in ESs was 78.83%. Studies mostly in the online repository (Table 3 in the online supplemental
involved typical students in Grades 6–12 in the United States (one materials).
study was in Canada and two studies involved students with special For the metaregressions examining if student grade level or time
needs). Eight comparisons applied an experimental design, and the of testing were associated with magnitude of ESs, a statistical detect-
most common controls were BAU, reading, and no treatments. able association between grade level and ESs was not found in any of
Students composed text in all studies. the 18 analyses conducted (all p . .05). Likewise, there were no stat-
Text Structure Instruction. Examples of such instruction istically detectable associations between posttest and maintenance
included teaching basic structural elements of persuasive text writing outcomes in the 14 types of writing instruction when this
(Scardamalia & Paris, 1985) and teaching elements for writing an could be tested (all p . .05).
essay in response to material read (Niemi et al., 2007). Such instruc-
tion produced statistically detectable ESs of 0.39 for all writing out-
Research Question 5: Does Teaching Writing Improve
comes (Table 2; 73% of ESs were positive) and 0.39 for writing Reading Outcomes?
quality (Table 4; 75% of ESs were positive). Variability in ESs When all reading outcomes were considered collectively, teaching
were both 89.78%. Studies mostly included typical Grade 6–12 stu- writing resulted in a statistically significant improvement in reading
dents in the United States (two studies involved students with special performance (ES = 0.22). Furthermore, teaching writing improved
needs, and two studies occurred in the Netherlands and one in reading comprehension by slightly more than one-fourth of a SD
Canada). All but one comparison applied an experimental design, (0.26). Statistically detectable improvements in reading achievement
and the most common control conditions were BAU, reading, and and reading vocabulary were not obtained.
diluted treatments. Students composed text in all comparisons.
Vocabulary Instruction. Examples of vocabulary instruction Research Question 6: Do Study Attributes Predict
included teaching new vocabulary related to the theme of students’ Magnitude of Reading Outcomes?
papers (Duin & Graves, 1987) and teaching students how to quickly
retrieve familiar words related to a writing topic (van Gelderen et al., There was considerable variability in effects for reading out-
2011). Vocabulary instruction did not produce statistically detect- comes, with 82.67% of the variance in ESs due to between-study
able results for all writing outcomes (see Table 2). Students in factors (Table 2). As can be seen in Table 3, none of the predic-
these studies were typical writers mostly in the United States in tors in the first two metaregressions accounted for statistically sig-
Grades 6–11 (one study occurred in the Netherlands). All compari- nificant variance in the magnitude of ESs. This included study
sons applied an experimental design, and the most common control quality indicators as well as study characteristics of grade level,
condition was BAU. Students composed in all but one condition. type of student, writing during treatment, and type of control con-
Increased Writing. Examples of increased writing included dition in the second metaregression. The covariate, time teaching
writing an additional essay each week (Lee & Schallert, 2015), writing, was also not statistically related to the magnitude of ESs
and writing descriptive text over 10, 50-min sessions (De Smedt for reading outcomes in 41 independent comparisons involving
et al., 2019). Such instruction did not result in ESs statistically 170 ESs in the third metaregression (average-weighted
greater than no effect for all writing outcomes or writing quality ES = −0.01).
 1022 GRAHAM ET AL.

Table 5 Figure 2
Moderation Analysis for Measures for Strategy Instruction and Funnel Plot
Digital Writing
Writing treatment and
outcomes K Estimate (SE) 95% CI I2
Strategy instruction 609 89.69
Intercept (writing quality) 127 0.87*** (.16) [.54, 1.19]
Ideation 92 −0.09 (.26) [−.62, .44]
Elements 69 −0.09 (.18) [−.47, .29]
Organization 88 −0.04 (.24) [−.54, .47]
Voice 12 −0.34 (.19) [−.88, .19]
Process 116 −0.12 (.24) [−.66, .41]
Translation 26 −0.53* (.23) [−1.06, .00]
Transcriptiona 21 0.20 (.85) [−2.37, 2.78]
−0.67* (.23) [−1.18, −.16]
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Motivational beliefs 58
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Digital writing 258 79.1

Intercept (writing quality) 61 0.31** (.10) [.10, .52]
Ideation 21 −0.04 (.18) [−.43, .35]
Elementsa 15 0.99 (1.20) [−3.68, 5.67]
Organization 12 −0.10 (.21) [−.58, .38]
Voice 9 −0.04 (.20) [−.56, .48] Likewise, there were improvements in the cognitive capabilities
Process 47 0.29 (.23) [−.25, .82] and resources writers draw on when composing. For example, teach-
Translation 49 −0.10 (.19) [−.51, .31]
Transcriptiona 22 −0.16 (.13) [−.60, .29]
ing writing enhanced all of the writing production processes identi-
Motivational beliefs 22 −0.46* (.17) [−.84, −.07] fied in the WWC model (Graham, 2018). Students evidenced
improved ideation, as represented by an increase the information
Note. K refers to the number of effect sizes; results are restricted to effect
sizes involving writing achievement only; reference group is writing
included in their text (ES = 0.46), enhanced translation, as revealed
quality. CI = confidence interval. through improvements in sentence construction skills (ES = 0.51),
a
df , 4; results should be taken with caution. better transcription skills, as indicated by better scores on measures
* p , .05. ** p , .01. *** p , .001. of spelling, handwriting, and typing (ES = 0.49), and strengthened
conceptualization and reconceptualization production processes, as
manifested in improvements in planning, revising, and self-regulation
Publication Bias processes captured by process outcome measures (ES = 0.82).
A funnel plot was created for all obtained outcomes (see Figure 2). Additionally, the LTM resources available to writers evidenced
The funnel plot appeared to be asymmetrical, with the possibility of improvements, as knowledge about writing increased (ES = 0.94)
missing ESs on the left side of the forest plot. The result from the and writing motivational beliefs became more positive (ES = 0.13).
Egger’s test was consistent with a possible concern about publica- It must be noted that some of these outcomes were assessed more fre-
tion bias because the intercept significantly deviated from zero quently than others in the studies reviewed here. Future studies need
(−1.78; p , .001). to more consistently assess writing organization, writing elements,
voice, writing processes, transcription skills, writing knowledge,
and motivational beliefs. None of these constructs were assessed in
Discussion more than 25% of the writing treatment/control comparisons.
A Pathway to Better Writing Is to Teach Writing There was considerable variability in the ESs for writing out-
comes. We examined if this variability was related to study features
When the ESs from the 406 independent writing treatment/con- (grade level, capability of writers, writing during treatment, assess-
trol comparisons from this meta-analysis were examined collec- ment timing, and type of control condition) and study quality indi-
tively, teaching writing resulted in almost one-half a SD cators (research design, publication type, N of 1 study, treatment
improvement in students’ writing (ES = 0.49). The average- fidelity, teacher effects, attrition greater than 20%, reliability of mea-
weighted ES for writing exceeded the overall ES of 0.28 reported sures, and norm-referenced assessments). It is important to examine
by Hillocks (1986) in his seminal meta-analysis with students in if such associations are statistically detectable because they can sug-
the elementary grades through college. There are multiple possible gest the presence of bias when obtained. With three exceptions, none
reasons for this difference. Hillocks’ focused on a broader range of of the predictors identified above were statistically related to the
grades than we did, there was a 10-fold increase in the number of magnitude of ESs for writing outcomes. One, the average-weighted
writing treatment/control comparisons since his earlier review, ES for writing treatment/control conditions published in journal arti-
and new very effective writing treatments were tested after his cles was 0.21 of a SD larger than the ES for comparisons presented in
review was conducted. other outlets. Consequently, it was important that we included both
The findings from our meta-analysis not only showed that writing journal articles and other types of publications in this review. Two,
instruction is valuable, it provided evidence supporting the theoret- the average-weighted ES for studies that were N of 1 was 0.20 of a
ical contributions of specific writing processes, skills, and resources SD larger than studies that were not N of 1. This suggests that our
to composing. In terms of what students wrote, overall writing qual- obtained ES of 0.49 for teaching writing may be biased by studies
ity improved (ES = 0.46) as did text organization (ES = 0.51), that were N of 1. As a result, we recommend a moratorium on con-
inclusion of genre elements (ES = 0.60), and voice (ES = 0.30). ducting N of 1 writing intervention investigations. In effect, they
 SECONDARY WRITING INSTRUCTION 1023

should not be a sanctioned approach for dissertations or theses, pub- Perin (2007). It presented the best match in terms of grade range
lished in journal articles, or viewed positively by promotion and ten- (4–12 to our 6–12). If a writing treatment examined in this review
ure committees. Three, norm-referenced tests produced smaller was not in Graham and Perin (2007), the next point of comparison
effects than researcher designed assessments (these latter tests was Graham and Harris (2018). Although this review covered
were 0.15 SD larger). Grades 1–12, it computed ESs for a greater range of writing treat-
We would further like to draw attention to the average-weighted ments than Graham and Perin (2007). As neither of the first two com-
ESs for writing outcomes administered just following treatment parisons calculated an overall ES for feedback, we consulted
(0.49 at posttest) and two or more weeks later (0.48 at maintenance), Graham et al. (2015) for such a statistic, despite the focus on
which was not statistically significant. In effect, there was no drop- Grades 1–8, because it provided the best available comparison. No
off in students’ writing outcomes when their performance was prior review calculated an ES for creativity/critical thinking.
assessed over time. This is a promising outcome, but it is important Outcomes for writing quality in this review were similar to those
to remember that maintenance was only assessed for 19% of the in Graham and Perin (2007) for writing strategy instruction (0.93 vs.
treatment/control comparisons conducted. We encourage research- 0.82) and prewriting activities (0.44 vs. 0.36). This was also the case
ers to more consistently collect maintenance data for writing (as for SRSD (1.16 vs. 1.14, respectively) and non-SRSD writing strat-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

well as reading outcomes) when testing the effects of teaching writ- egy instruction (0.79 vs. 0.69). Moreover, the average-weighted ES
ing to secondary students. This would be especially useful if main- for text structure instruction in this article (0.39) was similar to the
tenance was measured at multiple 3-month intervals. The success of outcome in Graham and Harris (2018), although the ES for writing
this recommendation rests, at least in part, on funding agencies plac- quality in the latter was not statistically greater than no effect (0.30).
ing greater emphasis on the collection of longitudinal follow-up Three of the writing treatments in the current review had larger
data. ESs for writing quality than were obtained by Graham and Perin
We were surprised that less than one-half of the studies reviewed (2007): emulating models of good writing (0.56 vs. 0.25), sentence
(45%) were published as journal articles. There were a surprisingly instruction (0.75 vs. 0.50), and grammar instruction (0.47 vs.
large number of studies that were unpublished dissertations and the- −0.32). Of these three writing treatments, the most remarkable dif-
ses (48%), raising questions about why these studies were not pub- ference involved grammar instruction. The most plausible reason for
lished in more accessible venues. Teacher effects were controlled in this turn-around was that Graham and Perin (2007) treated control
just over one-half (56%) of the studies reviewed, and treatment fidel- conditions that taught grammar as experimental conditions in their
ity was established in less than one-third of them. More positively, analysis. With the exception of one study, the obtained ESs in inves-
attrition over 20% was rarely a problem, and more than one-half tigations when this was done were all negative (ranging from −1.40
of the writing treatment/control comparisons (56%) applied an to −0.21). All other ESs for teaching grammar in their review were
experimental design. We need to do a better job of ensuring writing positive. Thus, eliminating studies when a control was treated as an
research is published in more accessible outlets, it controls for experimental condition would have resulted in a positive overall ES
teacher effects, and it consistently established treatment fidelity. for grammar instruction.
The average-weighted ESs for writing quality for goal setting
Secondary Students’ Writing Can Be Improved in Many (0.49) and peer assistance (0.43) in the present review were smaller
Different Ways than effects (0.70 and 0.75, respectively) in Graham and Perin
(2007). Similarly, the overall effects of providing feedback (ES =
We computed average-weighted ESs for 23 different writing treat- 0.27) was smaller than the effect (ES = 0.61) observed by Graham
ments. When we focused on all writing outcomes collectively, 16 of et al. (2015). Although effects for writing quality for teacher
these treatments produced statistically detectable effects. This (ES = 0.87), peer (ES = 0.51), self (ES = 0.62), and computer feed-
included comprehensive writing programs (ES = 0.47) as well as back (ES = 0.38) were statistically significant in Graham et al.
procedures to support students as they wrote: goal setting (ES = (2015), this was only the case for peer feedback in the current review
0.44), prewriting activities (ES = 0.49), inquiry (ES = 0.92), peer (ES = 0.53). Outcomes for digital writing were also less robust in
assistance (ES = 0.51), feedback (ES = 0.34), and the use of digital this review than in Graham and Harris (2018). The average-weighted
writing tools (ES = 0.31). It further included methods for enhancing ES of 0.15 and 0.26 for writing quality for word processing and
students’ knowledge of writing through emulating models of good word processing plus programs in this review were not statistically
writing (ES = 0.46) and observing writers and readers of writing different from no effect and were smaller than the statistically signif-
(ES = 0.41). Teaching writing was also effective as positive effects icant ESs of 0.44 for word processing and 1.46 for word processing
were obtained for teaching writing strategies (ES = 0.76), creativity/ plus in this prior review. There are many possible reasons for smaller
critical thinking (ES = 0.27), texts structure (ES = 0.39), sentence effects for these writing treatments, including a larger sample in this
instruction (ES = 0.73), grammar (ES = 0.77), transcription skills review from which to calculate more reliable effects (goal setting
(ES = 0.71), and summary writing (ES = 0.49). For 12 of these 16 and prewriting), differential effectiveness with younger and older
writing treatments, writing quality was also improved. We were students (e.g., feedback), and diminishing effects for writing treat-
unable to determine if this occurred for the other four writing treat- ments as they become more common or mundane (e.g., word pro-
ments (inquiry, observation, transcription instruction, and summary cessing and word processing plus).
writing instruction) because this construct was rarely assessed. Although the ES for writing quality for comprehensive writing
In comparing our findings for specific writing treatments to prior programs that did not involve the process approach was not statisti-
meta-analyses, we focused on writing quality because almost all pre- cally greater than no effect, it was 0.58, which is typically considered
vious reviews only calculated ESs for this construct. Our first point a practical effect. This lack of statistical significance was likely due
of comparison was a comprehensive meta-analysis by Graham and to considerable variability in writing quality outcomes for this
 1024 GRAHAM ET AL.

treatment (see Table 2). We were unable to compare this ES or the ranged from 0.22 to 0.62), which examined the effects of writing
one for process writing approaches affiliated with the NWP (ES = and writing instruction on the reading of Grade 1–12 students.
0.23) to findings from prior meta-analyses because such analyses One possible reason for this difference was that over one-half of
were not conducted. However, for process writing approaches not the studies in Graham and Hebert (57%) involved students in
affiliated with the NWP, the ES for writing quality in this review Grades 1–5, and writing instruction may be more effective in
(0.61) was larger than the average-weighted ES (0.32) for process improving younger students’ reading than it is in improving older
approach studies in Graham and Perin (2007). Nevertheless, the out- students’ reading. This assertion needs to be tested in future studies.
come for process writing in Graham and Perin was similar to the ES Even so, the current meta-analysis and the prior one by Graham and
of 0.42 obtained when a single effect for writing quality was calcu- Hebert (2011) demonstrated that teaching writing enhances stu-
lated for all process writing approach studies reviewed here (affili- dents’ reading, providing empirical evidence to support models of
ated with the NWP or not). reading and writing connections (see Shanahan, 2016). The finding
Importantly, some of the writing treatments included in this that writing can improve reading is especially important with
review were the subject of more research than others. For instance, secondary students, as reading is not commonly taught to them
strategy instruction, comprehensive writing programs, digital writ- (Biancarosa & Snow, 2004).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

ing, and feedback were tested 40 or more times each, but vocabulary When we examined the effects of teaching writing on reading,
instruction, imagery, inquiry, observation, revision instruction, and only one of the three reading outcomes was statistically significant.
enhancing motivation were tested less than 10 times each. Less con- The ES of 0.26 for reading comprehension was statistically greater
fidence can be placed in the reliability of findings when they are than no effect, but the ESs for reading achievement and vocabulary
based on a small number of studies. Even for writing treatments were not. Thus, gains in reading appear to be specifically accounted
tested more frequently, like strategy instruction, some types of strat- for by improvements in reading comprehension. Future writing treat-
egies (planning) have been tested more than others (revising), and ment studies need to more consistently assess a broader array of read-
the strategies tested are not equally distributed across genre or ing outcomes. In the current review, only eight of the writing
grades, making it difficult to determine their generalizability. treatment/control comparisons administered a reading achievement
Thus, there is a need for additional research to test the writing treat- measure and just 10 of them administered a vocabulary measure.
ments included in this review as well as test new writing treatments Word reading and reading fluency measures were administered so
which have not been the subject of any or much scientific testing. infrequently, we were unable to compute an average-weighted ES
The outcomes from this review provided support for multiple for either outcome.
aspects of the WWC model (Graham, 2018), which provided the Interestingly, there was considerable variability in ESs for
framework for this review. We assumed that if a writing treatment reading outcomes, but our predetermined moderators for study
was designed to improve a specific process or resource in the model quality indicators (e.g., research design, N of 1, publication type)
(e.g., translation skills), and there was a positive effect on writing over- and study characteristics (e.g., grade level) were not significantly
all, as measured by writing quality, this would provide support for that related to magnitude of effects. As more writing treatment studies
aspect of the model. Consequently, the outcomes for this meta- with reading outcomes become available, it may become more pos-
analysis provided support for the following writing production pro- sible to identify predictors that account for variability on reading
cesses: ideation (prewriting enhanced writing quality), translation outcomes.
(sentence and grammar instruction improved writing quality) and con-
ceptualization and reconceptualization (goal setting, teaching writing Implications for Education
strategies). The outcomes for writing strategy instruction also provided
support for students use of schemas (or strategies) as a tool to establish The findings from this meta-analysis have important implications
executive control over some aspect of the composing process. for classroom practice as well as policy recommendations. First, if
We further obtained support for the contention in the WWC teachers and policy-makers want to improve the writing of second-
model (Graham, 2018) that students draw on specialized knowledge ary students, then they need to do more than just increase how
about writing from LTM when composing (see Graham, 2018). This much time students spend writing. When all measures were consid-
was supported by our findings that studying and emulating models of ered collectively and when just writing quality was the focus of anal-
good writing and teaching text structure enhanced the quality of ysis, increasing time devoted to writing did not have a statistically
what students wrote. Finally, the hypothesized importance of social detectable impact (ES were 0.14 and 0.05, respectively). This does
processes in WWC model were supported by findings that feedback not mean that increasing how much students write has no positive
as well as having peers help each other resulted in qualitatively better impact. For example, writing more may introduce students to new
writing. As additional instructional research on writing emerges, ideas and content. We were unable to determine if this was the
evidence on additional aspects of the WWC and other writing mod- case because the studies reviewed did not assess such outcomes.
els will hopefully become available. Third, it is critical that teachers teach writing. Secondary students
became better writers when they were taught strategies for planning,
Teaching Writing Can Improve Reading drafting, revising, and editing; they learned to think about their writ-
ing more creatively and critically; and they were taught how to write
Outcomes for reading, which were included in 13% of the com- a good summary. Furthermore, teaching them how to construct more
parisons (52 comparisons), evidenced a small, but statistically sig- complex sentences and apply grammar correctly and effectively pro-
nificant, improvement of one-fifth of an SD. Our average-weighted vided them with needed writing skills. For some students, instruc-
ES of 0.22 for reading, however, was smaller than the ESs obtained tion to improve their spelling, handwriting, or keyboarding can
in Graham and Hebert’s (2011) meta-analysis (average-weighted ES also be advantageous.
 SECONDARY WRITING INSTRUCTION 1025

Third, it is clear that providing students with support as they write 12 documents involved Chinese. As additional studies accumulate
is beneficial. This included helping students access and organize in the future, it should be possible to determine if the effects of spe-
content for writing through prewriting activities and inquiry. It fur- cific writing instructional treatments differ from one writing system
ther included making clear the aims of writing through goal setting to the next.
as well as providing students with feedback and having them assist Like meta-analysts before us (e.g., Hillocks, 1986), we made a
each other as they compose. host of decisions about what studies to include, what constituted a
Fourth, it is profitable to put into place procedures that help sec- writing treatment, and how to group these treatments into coherent
ondary students learn more about what constitutes good writing. instructional categories. Although we specified how we did this
This includes providing them with good models of writing to emu- and provided data to show we did it reliably, we have no doubt others
late; asking them to observe others as they write or react to others’ will questions one or more of our decisions.
writing, and teaching them about the purposes and construction of Finally, 48% of the documents included in this review involved
different types of text. dissertations and thesis and another 7% included research reports,
Fifth, secondary students write better when using digital tools. ERIC documents, chapters, unpublished studies from authors, and
This included using word processing as a medium for composing conference papers. The studies included in these documents were
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

not subjected to the peer-review process. Based on our finding that

This document is copyrighted by the American Psychological Association or one of its allied publishers.

in addition to using digital writing programs that provided tools

that assisted writers or allowed them to compose multimodal text. studies in journal articles evidenced larger effects than nonjournal
It also included the use of computer-assisted instruction as a publications, including just peer-reviewed articles however would
means for teaching writing. have overestimated the effects of writing instruction with secondary
Finally, if teachers and policy-makers want to enhance secondary students. Even so, a best evidence meta-analysis that focuses just on
students’ reading, then writing instruction needs to become a consis- high-quality peer-reviewed and non-peer-reviewed studies would
tent and valued part of middle and high school students’ education. provide a needed addition to the literature. Such an analysis might
Currently, students do little writing at school and little time is spent exclude N of 1studies, investigations with high attrition, ESs based
teaching them to write (Graham & Rijlaarsdam, 2016). This needs to on unreliable measures, and so forth.
change if the power of writing instruction as a catalyst to improve
secondary students’ reading is to be realized. Concluding Comments
The meta-analysis reported here demonstrated that teaching
Limitations
writing to secondary students improved both their writing and
Although the current review included 406 independent writing reading. Just as importantly, we identified a variety of different
treatment/control comparisons and multiple procedures were writing treatments that can improve how these students write.
enacted to conduct a thorough search, examination of a funnel plot As a result, the review provides fertile information that schools
containing all ESs and outcome of the Egger’s test suggests that pub- and teachers can use to improve writing and policy-makers can
lication bias may be evident in this meta-analysis. It is important to apply when considering the importance of writing and how it
keep this in mind when interpreting the results. Because missing should be taught.
studies were most likely from the left side of the funnel plot, it is pos-
sible that the overall effect reported here may be somewhat inflated. References
In this review, we only examined specific types of quantitative
studies (experimental and quasi-experimental investigations). References for the studies included in the meta-analysis are found in online
supplemental materials.
Other types of quantitative research such as single participant
designs also provide important information about the effectiveness Andrews, R., Torgerson, C., Beverton, S., Freeman, A., Locke, T., Low, G.,
of writing treatments (see Rogers & Graham, 2008). Likewise, qual- Robinson, A., & Zhu, D. (2006). The effects of grammar teaching on writ-
itative studies provide valuable insights into how writing treatments ing development. British Educational Research Journal, 32(1), 39–55.
https://doi.org/10.1080/01411920500401997
are applied by teachers and received by students as well as how writ-
Applebee, A., & Langer, J. (2011). A snapshot of writing instruction in mid-
ing is taught by exceptional teachers. To obtain a more complete pic-
dle schools and high schools. English Journal, 100(6), 14–27. https://lead
ture of effective practices for teaching writing to secondary students, .nwp.org/wp-content/uploads/2017/05/snapshot.pdf
the findings reported in this review need to be considered along with Biancarosa, G., & Snow, C. (2004). Reading next—A vision for action and
outcomes form other reviews that employ different methodologies. research in middle and high school literacy: A report to the Carnegie cor-
Many of the studies in this review did not provide information on poration of New York. Alliance for Excellence in Education.
locale (e.g., urban, SES, or race of study participants. When such Bloom, B., Engelhart, M., Furst, E., Hill, W., & Krathwohl, D. (1956).
information was provided for these variables as well as for other Taxonomy of educational objectives: The classification of educational
characteristics such as gender, separate pretest, posttest, and mainte- goals. Handbook I: Cognitive domain. David McKay Company.
nance data by treatment and control conditions were rarely provided Borrelli, B. (2011). The assessment, monitoring, and enhancement of treatment
fidelity in public health clinical trials. Journal of Public Health Dentistry,
for any of these attributes. Thus, it was not possible to determine if
71(1), S52–S63. https://doi.org/10.1111/j.1752-7325.2011.00233.x
such student characteristics were related to magnitude of effects.
Carroll, J. B. (1963). A model for school learning. Teachers College Record:
Even though most studies included in this meta-analysis involved The Voice of Scholarship in Education, 64(8), 1–9. https://doi.org/10
writing treatments aimed at the English language, 18% of the docu- .1177/016146816306400801
ments tested writing treatments that focused on written languages Cheung, A., & Slavin, R. (2016). How methodological features affect effect
other than English. The writing treatments in most of these addi- sizes in education. Educational Researcher, 45(5), 283–292. https://
tional documents (81%) centered on an alphabetic language, but doi.org/10.3102/0013189X16656615
 1026 GRAHAM ET AL.

Cortina, J., & Nouri, H. (2000). Effect size for ANOVA design. Sage. Hopewell, S., Loudon, K., Clarke, M. J., Oxman, A. D., & Dickersin, K.
Couzijn, M. (1999). Learning to write by observation of writing and reading (2009). Publication bias in clinical trials due to statistical significance or
processes: Effects on learning and transfer. Learning and Instruction, 9(2), direction of trial results. Cochrane Database of Systematic Reviews,
109–142. https://doi.org/10.1016/S0959-4752(98)00040-1 2009(1), Article MR000006. https://doi.org/10.1002/14651858.MR000006
Drew, S., Olinghouse, N., Faggella-Luby, M., & Welsh, M. (2017). .pub3
Framework for disciplinary writing in science Grades 6–12: A national Kim, Y.-S. G. (2020). Interactive dynamic literacy model: An integrative the-
survey. Journal of Educational Psychology, 109(7), 935–955. https:// oretical framework for reading and writing relations. In R. Alves,
doi.org/10.1037/edu0000186 T. Limpo, & M. Joshi (Eds.), Reading-writing connections: Towards inte-
Elma, C., & Bütün, E. (2015). Teachers opinions about primary and second- grative literacy science (pp. 11–34). Springer. https://doi.org/10.1007/
ary school students writing skills. Abant İ zzet Baysal University Journal of 978-3-030-38811-9_2
Education Faculty, 15(2), 104–131. Kim, Y.-S. G. (2022). Co-occurrence of reading and writing difficulties: The
Fisher, Z., Tipton, E., & Zhipeng, H. (2017). Package “robumeta” application of the interactive dynamic literacy model. Journal of Learning
[Computer software]. http://cran.uni-muenster.de/web/packages/robumeta/ Disabilities, 55(6), 447–464. https://doi.org/10.1177/00222194211060868
robumeta.pdf Koster, M., Tribushinina, E., de Jong, P. F., & van den Bergh, H. (2015).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Fitzgerald, J., & Shanahan, T. (2000). Reading and writing relations and their Teaching children to write: A meta-analysis of writing intervention
This document is copyrighted by the American Psychological Association or one of its allied publishers.

development. Educational Psychologist, 35(1), 39–50. https://doi.org/10 research. Journal of Writing Research, 7(2), 249–274. https://doi.org/10
.1207/S15326985EP3501_5 .17239/jowr-2015.07.02.2
Goldberg, A., Russell, M., & Cook, A. (2003). The effect of computers on Morphy, P., & Graham, S. (2012). Word processing programs and weaker
student writing: A meta-analysis of studies from 1992 to 2002. The writers/readers: A meta-analysis of research findings. Reading and
Journal of Technology, Learning, and Assessment, 2(1). https://ejournals Writing, 25(3), 641–678. https://doi.org/10.1007/s11145-010-9292-5
.bc.edu/index.php/jtla/article/view/1661 National Center for Educational Statistics. (2012). The nation’s report card:
Graham, S. (2018). The writer(s)-within-community model of writing. Writing 2011 (NCES 2012-470). Institute of Educational Sciences, U.S.
Educational Psychologist, 53(4), 258–279. https://doi.org/10.1080/ Department of Education. https://nces.ed.gov/nationsreportcard/pdf/
00461520.2018.1481406 main2011/2012470.pdf
Graham, S., & Harris, K. R. (2014). Conducting high quality writing inter- National Commission on Writing. (2003). The neglected “R”: The need for a
vention research: Twelve recommendations. Journal of Writing writing revolution. College Board.
Research, 6(2), 89–123. https://doi.org/10.17239/jowr-2014.06.02.1 Nunnally, J. (2017). Psychometric theory. McGraw-Hill.
Graham, S., & Harris, K. R. (2018). Evidence-based writing practices: A Nye, B., Konstantopoulos, S., & Hedges, L. (2004). How large are teacher
meta-analysis of existing meta-analyses. In R. Fidalgo, K. R. Harris, & effects? Educational Evaluation & Policy Analysis, 26(3), 237–257.
M. Braaksma (Eds.), Design principles for teaching effective writing: https://doi.org/10.3102/01623737026003237
Theoretical and empirical grounded principles (pp. 13–37). Brill Pressley, M., Graham, S., & Harris, K. R. (2006). The state of educational
Editions. intervention research. British Journal of Educational Psychology, 76(Pt
Graham, S., & Hebert, M. (2011). Writing-to-read: A meta-analysis of the 1), 1–19. https://doi.org/10.1348/000709905X66035
impact of writing and writing instruction on reading. Harvard R Core Team. (2018). R: A language and environment for statistical comput-
Educational Review, 81(4), 710–744. https://doi.org/10.17763/haer.81.4 ing [Computer software]. https://www.r-project.org/
.t2k0m13756113566 Rogers, L., & Graham, S. (2008). A meta-analysis of single subject design
Graham, S., Hebert, M., & Harris, K. R. (2015). Formative assessment and writing intervention research. Journal of Educational Psychology,
writing: A meta-analysis. Elementary School Journal, 115(4), 524–547. 100(4), 879–906. https://doi.org/10.1037/0022-0663.100.4.879
https://doi.org/10.1086/681947 RStudio Team. (2016). RStudio: Integrated development for R [Computer
Graham, S., & Perin, D. (2007). A meta-analysis of writing instruction for software]. http://www.rstudio.com/
adolescent students. Journal of Educational Psychology, 99(3), 445– Schulz, K., & Grimes, D. (2002). Sample size slippages in randomized trials:
476. https://doi.org/10.1037/0022-0663.99.3.445 Exclusions and the lost and wayward. The Lancet, 359(9308), 781–785.
Graham, S., & Rijlaarsdam, G. (2016). Writing education around the globe: https://doi.org/10.1016/S0140-6736(02)07882-0
Introduction and call for a new global analysis. Reading & Writing, 29, Shanahan, T. (2016). Relationships between reading and writing develop-
781–792. https://doi.org/10.1007/s11145-016-9640-1 ment. In C. A. MacArthur, S. Graham, & J. Fitzgerald (Eds.), Handbook
Graham, S., Tavsanli, O., & Kaldirim, A. (2022). Improving writing skills of of writing research (2nd ed., pp. 194–207). Guilford.
students in Turkey: A meta-analysis of writing interventions. Educational Slavin, R. E., & Madden, N. A. (2011). Measures inherent to treatments in
Psychology Review, 34(2), 889–934. https://doi.org/10.1007/s10648-021- program effectiveness reviews. Journal of Research on Educational
09639-0 Effectiveness, 4(4), 370–380. https://doi.org/10.1080/19345747.2011
Harris, K. R., & Graham, S. (1992). Self-regulated strategy development: A .558986
part of the writing process. In M. Pressley, K. R. Harris, & J. Guthrie (Eds.), Suggate, S. (2016). A meta-analysis of the long-term effects of phonemic
Promoting academic competence and literacy in school (pp. 277–309). awareness, phonics, fluency, and reading comprehension interventions.
Academic Press. Journal of Learning Disabilities, 49(1), 77–96. https://doi.org/10.1177/
Hedges, L. V., Tipton, E., & Johnson, M. C. (2010). Robust variance estima- 0022219414528540
tion in meta-regression with dependent effect size estimates. Research Swanson, L., Harris, K. R., & Graham, S. (2013). Handbook of learning dis-
Synthesis Methods, 1(1), 39–65. https://doi.org/10.1002/jrsm.5 abilities (2nd ed.). Guilford.
Higgins, J. P. T., & Green, S. (Eds.). (2011). Cochrane handbook for system- Tanner-Smith, E. E., Tipton, E., & Polanin, J. R. (2016). Handling complex
atic reviews of interventions. http://handbook.cochrane.org meta-analytic data structures using robust variance estimates: A tutorial in
Hillocks, G. J. (1986). Research on written composition: New directions for R. Journal of Developmental and Life-Course Criminology, 2(1), 85–112.
teaching. National Council of Teachers of English. https://doi.org/10.1007/s40865-016-0026-5
Hillocks, G. J. (2008). Writing in secondary schools. In C. Bazerman (Ed.), Tipton, E. (2013). Robust variance estimation in meta-regression with binary
Handbook of research on writing: History, society, school, individual, text dependent effects. Research Synthesis Methods, 4(2), 169–187. https://
(pp. 311–329). Routledge. doi.org/10.1002/jrsm.1070
 SECONDARY WRITING INSTRUCTION 1027

Tipton, E., & Pustejovsky, J. E. (2015). Small-sample adjustments for tests of Troia, G. (2014). Evidence-based practices for writing. CEEDAR Center.
moderators and model fit using robust variance estimation in meta- https://ceedar.education.ufl.edu/wp-content/uploads/2014/09/IC-5_FINAL_
regression. Journal of Educational and Behavioral Statistics, 40(6), 08-31-14.pdf
604–634. https://doi.org/10.3102/1076998615606099 Tukey, J. W. (1977). Exploratory data analysis. Addison-Wesley.
Tolchinsky, L. (2016). From text to language and back again: The emergence
of written language. In C. A. MacArthur, S. Graham, & J. Fitzgerald Received January 28, 2023
(Eds.), Handbook of writing research (2nd ed., pp. 144–159). Guilford Revision received June 14, 2023
Press. Accepted June 23, 2023 ▪
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.