Reading Research for Students

with LD :
A Meta-Analysis of Intervention Outcomes
H. Lee Swanson

The present article provides a meta-analysis of instructional research with samples of children and adolescents with learning
disabilities in the domains of word recognition and reading comprehension. The results of the synthesis showed that a prototypical
intervention study has an effect size (ES) of .59 for word recognition and .72 for reading comprehension. Four important findings
emerged from the synthesis: (a) Effect sizes for measures of comprehension were higher when studies included derivatives of both
cognitive and direct instruction, whereas effect sizes were higher for word recognition when studies included direct instruction;
(b) effect sizes related to reading comprehension were more susceptible to methodological variation than studies of word recognition;
(c) the magnitude of ES for word recognition studies was significantly related to samples defined by cutoff scores (IQ > 85 and
reading < 25th percentile), whereas the magnitude of ES for reading comprehension studies was sensitive to discrepancies between
IQ and reading when compared to competing definitional criteria; and (d) instructional components related to word segmentation
did not enter significantly into a weighted least square hierarchical regression analysis for predicting ES estimates of word recognition
beyond an instructional core model, whereas small-group interactive instruction and strategy cuing contributed significant variance
beyond a core model to ES estimates of reading comprehension. Implications related to definition and instructional components that
optimize the magnitude of outcomes are discussed.

eading difficulty is one of the to some of these questions, the pur- studies focus on phonological aware-
most significantchildren
JL
problems
identi-
ex- pose of this article was to synthesize
the empirical evidence derived from
ness, or the
ability to code words
perienced by intoindividually assigned units (e.g.,
fied with learning disabilities (LD). research on reading intervention for Brown & Felton, 1990; Jones, Torgesen,
Reading underlies their performance students with LD. & Sexton, 1987; Kennedy & Backman,
in most academic domains (e.g., There are two reasons for the cur- 1993; Vellutino et al., 1996). From these
Berninger, 1994; Bishop & Adams, rent synthesis, one practical and the data the case has been made that a
1990; Bruck, 1990; Ehri, 1994; Guthrie, other theoretical. In terms of practice, revision in teaching methods is called
VanMeter, McMann, & Wigfield, 1996), because of the diversity of processing for nationwide in which current
as well as their adjustments to most difficulties attributed to children with context-based reading instruction is re-
school activities (e.g., Byrne, Freebody, LD (e.g., Borkowski, Weyhing, & Carr, placed with highly structured, explicit,
& Gates, 1992; Share & Stanovich, 1995; 1988; Olson, Wise, Johnson, & Ring, and intensive instruction in phonologi-
Stanovich, 1986). Over the last 30 years 1997; Shankweiler & Crain, 1986), there cal rules and other applications to
there have been several conceptual are divisions about the most effec- print.
shifts, as well as paradigm debates, tive method of teaching reading (see On the other side of the continuum
regarding what underlies reading Adams & Bruck, 1993; Foorman, 1994; are those who suggest that reading
problems in children with LD, which Palincsar, 1986; Palincsar & Brown, has meaning within the context of lan-
in turn raised questions about the best 1984; Pressley & Rankin, 1994; and guage (see Adams & Bruck, 1993;
instructional intervention for re- Vellutino et al., 1996, for a review). Foorman, 1994; Frederiksen & War-
mediating such problems. In response On one side of the continuum, several ren, 1987; and Pressley & Rankin, 1994,

Note. This work is a subsection of a final report submitted to the U.S. Department of Education. The complete report was published
in Interventions for Students with Learning Disabilities: A Meta-Analysis of Treatment Outcomes, by H. L. Swanson, with M. Hoskyn and
C. Lee, 1999, New York: Guilford Press. Copyright 1999 by Guilford Press. Adapted with permission.
 505

for a comprehensive review of these field because reading problems are ing measures. Strategy and direct in-
diverse orientations). It is assumed that pervasive in populations with LD, and struction models have yielded high
(a) research showing generalization of there is a plethora of research sug- effect sizes in our previous syntheses
isolated-word interventions to read- gesting that phonological coding defi- (Swanson, Carson, & Sachse-Lee, 1996;
ing fluency and comprehension is lim- cits underlie these difficulties, partic- Swanson & Hoskyn, in press) and
ited (e.g., Aaron, Frantz, & Manges, ularly in sight word recognition (see therefore were compared to other
1990; Kendall & Hood, 1979; Pflaum, Siegel, 1993, for a review). Unfortu- models in the present synthesis. The
1980; then see Byrne et al., 1992), and nately, there have been no comprehen- focus of these two models is also in-
(b) the primary function of reading is sive syntheses of intervention studies fluenced by recent studies that have
extracting meaning from text (Palin- to determine how responsive word assessed these two approaches in word
csar & Brown, 1984; Pinnell, Lyons, recognition is to intervention. That recognition studies (e.g., Lovett et al.,
DeFord, Bryk, & Seltzer, 1994). Instruc- is, on the basis of the assumption 1994) and reading comprehension
tion is anchored on the premise that that LD results from an information- studies (e.g., Graves, 1986). Thus, we
there is a strong link between reading processing deficit with a neurological tested whether treatments that include
acquisition and oral language acqui- base, and that such a deficit is not due components of strategy instruction (SI),
sition and that when reading is taught, to faulty instruction, one would pre- direct instruction (DI), or a combina-
its meaning and purpose should be dict some resistance (a generally low tion of both yield higher effect sizes
emphasized. Instruction is geared or slow degree of change or success) than competing models.
toward providing students with the in a core deficit area. That is, instruc- Second, we tested whether certain
information and metacognitive skills tion must be intense (one-to-one), be models of instruction (e.g., SI mod-
necessary for academic success explicit, and occur over an extended els) have broad effects across word-
through the use of self-monitoring period of time to allow for generaliza- recognition and comprehension mea-
and interactive dialogue (e.g., Bos & tion. Unfortunately, this assumption sures or are specific to one domain.

Anders, 1990; Johnson, Graham, has not been tested. One qualitative This issue is difficult to address in in-
& Harris, 1997; Palincsar & Brown, criterion for measuring resistance to dividual reading intervention studies
1984, 1988; Pressley, Brown, El-Dinary, change is the magnitude of ES between because, in some cases, outcome mea-
& Allferbach, 1995; Wong, 1991; Wong the control and experimental condi- sures are highly similar to treatment
& Jones, 1982). In summary, although tions. According to Cohen (1988), effect activities (see Swanson & Hoskyn,
practitioners may incorporate the ef- sizes above .80 are considered sub- 1998, and Weisz, Weiss, Han, Granger,
fective aspects of each approach into stantial, above .60 are considered & Morton, 1995, for a discussion of
their instruction, there remains some moderate, and below .20 are consid- this issue) and therefore confound
debate as to whether instruction ered minimal. Based on Cohen’s cri- conclusions that can be drawn about
should primarily emphasize word teria, one would not necessarily expect the independent effects of treatment.
skills or meaning (via metacognitive substantial changes in a core deficit For example, when one treats a child’s
skills), or emphasize both simulta- (e.g., ES would be expected to hover word recognition problems with phon-
neously. around .20) in a short period of time. ics instruction, the most valid test of
In terms of theory, a conceptual One would expect, however, more sub- the child’s ability to read is a test
model that has some consensus in stantial changes the longer explicit in- of real-word recognition and not per-
the field is that children with LD are struction in reading is implemented (see formance on pseudowords or phonics
of normal intelligence but suffer Note 1). measures. Likewise, when one treats
information-processing difficulties The purpose of this article was to a child’s reading comprehension prob-
(e.g., Borkowski, Estrada, Milstead, & synthesize the intervention literature, lems with metacognitive training, the
Hale, 1989; Deshler & Shumaker, 1988; via a meta-analysis, focusing on stud- ability to comprehend text, rather than
Fletcher et al., 1994; Stanovich & Siegel, ies that included measures of word responses metacognitive ques-
on a

1994; Swanson & Alexander, 1997). A recognition and reading comprehen- tionnaire, is probably the most valid
popular assumption that has emerged sion. In this manner we could com- test of training. Thus, isolating the
in the last few years is that children pare studies that directly focused on specific effects of treatment requires
with learning disabilities have specific reading and those that indirectly in- control of confounds related to the
processing deficits that are localized fluenced reading. In addition to char- dependency between the domain (de-
in phonological processing, particu- acterizing the literature along several pendent measure) of the estimated ES
larly at the word-recognition level descriptive dimensions (e.g., instruc- and the treatment (independent vari-
(e.g., Francis, S. Shaywitz, Stuebing, tional and methodological procedures), able).
B. Shaywitz, & Fletcher, 1996; Siegel, three additional issues were of inter- To address this issue, the present
1992; Stanovich & Siegel, 1994). This est. First, we compared the effects of synthesis used as the outcome mea-
assumption finds a consensus in the general instructional models on read- sure those variables considered appro-
506

priate for transfer (i.e., word recog- more general or less specific defini- dition, every state department (and
nition and comprehension). The tions of LD. That is, we assume that director) of education was sent a let-
difficulty in this comparison, however, the face validity of a definition is en- terrequesting technical reports on in-
is that some studies include both mea- hanced if one can show that (a) such a tervention studies for children and
sures and therefore are highly corre- definition is significantly related to adolescents with LD.
lated with the treatment variable. A treatment outcomes and (b) the mag- The pool of relevant literature was
weighted least squares (WLS) regres- nitude of these outcomes is signifi- narrowed down to studies that uti-
sion has been suggested as one ap- cantly different from those in studies lized an experimental design in which
proach to handle a wide variety of with less specific definitions. children or adults with LD received
these multivariate statistical problems treatment to enhance their academic
(Gleser & Olkin, 1994) and was uti- performance. This procedure nar-

lized in this analysis. Method rowed the search to 913 data-based

Finally, we tested the assumption articles (or reports) that focused on
that general aptitude measures mod- various domains (reading, mathemat-
Data Collection
erate treatment outcomes. Reading ics, writing). After a review of these
scores at or below the 25th percentile The PsycINFO, MEDline, and ERIC studies, each data-based report was
in reading recognition (see Fletcher online data bases were systematically evaluated on the following six addi-
et al., 1994; Siegel & Ryan, 1989; scanned for studies from 1963 to 1997 tional criteria for study inclusion:
Stanovich & Siegel, 1994) and stan- (July 1) that met the inclusion criteria
dardized intelligence scores at or described below. The computer search 1. The study included a measure of
above 85 have been considered criti- strategy used the following terms: real word recognition and / or
cal cutoff points for defining LD (see learning disabled (disabilities), reading reading comprehension. Thus, the
Morrison & Siegel, 1991, and Swanson, disabled (disabilities), dyslexic, education- focus of intervention may not
1991, for a discussion of this issue). ally handicapped, and slow learners, have been reading (e.g., spelling
The rise of cutoff scores in the experi- paired with variations of intervention, instruction may be emphasized),
mental literature has been in response treatment, training, remediation or in- but outcomes of treatment must
to (a) poorly defined samples in the struction. This search yielded approxi- have included reading measures
public schools and / or (b) the poor mately 2,900 abstracts, from articles, (see Note 2). ~ .
.

discriminant validity of discrepancy- technical reports, chapters, and dis- 2. The study included at least one
defined children (e.g., IQ scores are sertations. We examined all the ab- between-instructions comparison
generally 15 standard score points stracts prior to study selection to condition (i.e., control condition)
higher than reading scores; however, eliminate those studies that clearly did or within-design control condi-
see Reynolds, 1981) from generally not meet the inclusion criteria (e.g.,
tion (e.g., repeated measures
low-achieving children (i.e., children articles were reviews or position pa-
with both
design) that included participants
low-average IQ and low pers). Because the computer search with LD. Thus, studies that
reading scores) on measures of ortho- procedures excluded unpublished included only a pretest and
graphic, phonological, and memory studies and the most recent literature,
posttest, without an instructional
measures (Fletcher et al., 1994; Siegel, researchers (as identified by journal control condition of participants
1992). The implication of the latter board affiliations with the Learning with LD, were excluded.
findings is that IQ level may be irrel- Disability Quarterly, Journal of Learn- 3. The
evant to processing outcomes when ing Disabilities, and Learning Disabili- study provided sufficient
groups are generally poor readers (e.g., ties Research and Practice and / or mem- quantitative information to
Stanovich & Siegel, 1994). It has been bership in the International Academy permit the calculation of effect
sizes. These were calculated from
suggested, therefore, that children with for Research in Learning Disabilities) the means and standard devia-
LD be operationally defined in terms were sent letters requesting copies of
of average IQs and reading scores, es- unpublished and / or ongoing interven- tions of the performance out-
comes for the experimental and
pecially on word recognition mea- tion studies. We also hand searched control conditions, or from tests
sures, below the 25th (standard score the following journals for articles that
of 90) percentile (also see Foorman, did not emerge from the computer of the significance of the differ-
ences in performance between the
Francis, Fletcher, & Lynn, 1996, and search: Journal of Learning Disabilities,
Morrison & Siegel, 1991, for a discus- Journal of Educational Psychology, Learn- instruction conditions (e.g., t or
F tests, x2, exact p values).
sion). The treatment validity of such a ing Disability Quarterly, Reading and
cutoff-score definition, however, has Writing, Learning Disabilities Research 4. The recipients of the intervention
not been tested as a function of treat- and Practice, Exceptional Children, and were identified as children or
ment outcomes when compared to The Journal of Special Education. In ad- adults having average intelligence
 507

but experiencing problems in a ing may have provided measures of component measures of word recog-
particular academic, social, or process (e.g., metacognition) but failed nition, such as phonics, and nonword
related behavior domain. The to provide measures of transfer to real- [pseudoword] recognition) and meta-
mean IQ score reported for the word recognition and comprehension. cognitive skills (this domain included
LD participant sample could be measures of self-monitoring, speed of

derived from any standardized processing, problem-solving processes,

verbal or nonverbal intelligence Coding and Interrater synthesis, problem-solving strategies,
test or subtest. The cutoff score Agreement and awareness of strategy use).
for inclusion in the synthesis was For the subsequent statistical analy-
The general categories of coding
a reported meanstandardized IQ ses, only one ES (average ES within
included: (a) sample characteristics,
score > 84. Studies that reported each study) as a function of each do-
no IQ score but stated that the
(b) demographics, (c) quality of re- main category (e.g., word recognition)
search methodology, (d) components
group IQ scores were in the and
was computed, thereby partially con-
parameters of instruction,
average range were included in
(e) generalization / transfer, and (f) con- trolling for statistical independence
the analysis. If a study did not within the meta-analysis (see Gleser
ditions of treatment. Interrater agree-
state that IQ was in the average & Olkin, 1994, for a discussion). That
ments across various items exceeded
range, it was not included in the is, if a study had multiple dependent
80% in some areas (instructional com-
general analysis (see Note 3). measures within the reading recogni-

5. The treatment group received ponents, key terms) and 95% in others tion and
(dependent measures, statistical infor-
comprehension category,
instruction or assistance that was those that assessed the same constructs
above and beyond what they mation). Interrater agreement was also were combined and those that assessed
obtained on rejected articles. Four
would have received during the different constructs were reported in
doctoral students served as indepen-
course of their typical classroom the appropriate categories.
dent raters on the coding of each re-
experience. That is, the study
focused on treatment, rather than port. Raters were trained in the coding
merely a description of the child’s
of each category. Coding training was Categorization of Treatment
conducted during the first several Variables
current placement followed by an
months of this project; during that
evaluation. The fine line between Forty-five instructional activities (see
time, code definitions with low inter-
intervention and experimentation
rater agreement were refined, and rat- examples used to form the 20 instruc-
was drawn by requiring that the tional components below) were coded
ers were provided feedback for coding.
training or remediation was as present or not present in the study.
Approximate time required to code On the basis of a set number of activi-
dispensed over a minimum of each article varied from 2 to 3 hours.
3 days.
Interrater agreement on the dependent
ties, studies were classified into one
6. The study had to have been of four models: combined SI and DI
measures was calculated by using the
written in English. model (referred to as the combined
formula agreements divided by the
number of agreements plus disagree- model), DI alone, SI alone, and a model
Ninety-two studies were finally se- ments, multiplied by 100. Interrater
(non-SI and non-DI) that failed to reach
lected for analyses (see the Appen- a critical threshold of &dquo;reported&dquo; in-
agreements below 80% called for a
dix). Interrater agreement for article redefinition of the particular code and
formation.
inclusion and exclusion exceeded 90%. Based on several literature reviews
the retraining of raters relevant to in-
Although design issues (no control formation in each report. (e.g., Engelmann & Carnine, 1982;
condition) were the most frequent cri- Kameenui, Jitendra, & Darch, 1995;
teria for article exclusion, the inability Slavin, Karweit, & Madden, 1989;
to calculate ES, lack of clarity about
Categorization of Dependent Spector, 1995), those activities coded
whether students with LD were in- Measures that reflected direct instruction in the
cluded (e.g., IQ scores < 84, or aver- present synthesis were as follows:
age IQ wasnot stated); the inability to Dependent measures were coded into (a) breaking down a task into steps,
separate the performance of students one of two general categories as fol- (b) administering probes, (c) adminis-
with LD from that of other ability lows : (1) The word recognition domain tering feedback repeatedly, (d) provid-
groups; no information on sample size; included all measures of real-word ing a pictorial or diagram presentation,
and / or faulty statistical applications recognition, and (2) the reading com- (e) allowing for independent practice
(e.g., incorrect degrees of freedom) prehension domain included all mea- and individually paced instruction,
were also frequent reasons for article sures related to silent or oral reading (f) breaking the instruction down
exclusion. Another vexing situation comprehension. We also coded mea- into simpler phases, (g) instructing
was that studies that focused on read- sures of word skills (this included all in a small group, (h) teacher model-
508

ing a skill, (i) providing set materials (e) verbal-interactive dialogue-teacher student, sequencing short activi-
at a rapid pace, (j) providing indi- and student talk back and forth; (f) pro- ties, and using step by step
vidual child instruction, (k) teacher cess questions asked by teacher; and prompts;
asking questions, and (1) teacher pre- (g) assistance provided by teacher only 2. Drill / repetition and practice /
senting the novel materials. Any study as necessary. review-statements in the treat-
that included a minimum of four of Studies were then classified into one ment description related to
these codes in the treatment phase was of four models: SI+DI (referred to as mastery criteria, distributed
labeled a direct instruction. the Combined Model), DI alone, SI alone, review and practice, use of
A second instructional variable and the Non-SI+Non-DI model. As a redundant materials or text,
coded was strategy instruction. Com- validity check on our classifications, repeated practice, sequenced
ponents related to effective strategy we compared our classification of the reviews, daily feedback, and
instructional programs are reviewed treatment conditions with that of the weekly reviews;
elsewhere (see Borkowski & Turner, primary author’s general theoretical 3. Anticipatory or preparation
1990, and Pressley & Ghatala, 1990, model and/or the label attached to responses-statements in the
for a review). Some of these compo- the treatment condition. There was treatment description related to
nents include advance organizers (pro- substantial overlap (approximately asking the child to look over
vide students with a type of mental 70% of the studies) between those stud- material prior to instruction,
scaffolding in which to build new ies we classified as DI and SI models directing the child to focus on
understanding, i.e., consist of infor- and the primary authors’ description material or concepts prior to
mation already in the students’ minds of the independent variables. For ex- instruction, providing informa-
and the new concepts that can orga- ample, frequent terms provided by the tion to prepare student for
nize this information); help with orga- author were strategy, cognitive inter- discussion, and stating the
nization (questions directed at students vention, monitoring, metacognition, self- learning objective for the lesson
from time to time to assess their under- instruction, and cognitive-behavior prior to instruction;
standing) ; strategies for elaboration modification for the SI model. Those 4. Structured verbal teacher-student
(thinking about the material to be that were classified as DI by our crite- interaction-statements in the
learned in a way that connects the ria used such labels as directed instruc- treatment description about
material to information or ideas al- tion, advance organizers, adapting mate- elaborate or redundant explana-
ready in their minds); generative learn- rials, corrective feedback, and decoding. tions, systematic prompting of
ing (learners must work to make sense We were also interested in identify- students to ask questions,
out of what they are learning by sum- ing an instructional core that influ- teacher-student dialogue, and
marizing the information); general enced treatment outcomes indepen- open-ended or directed questions
study strategies (e.g., underlining, note dent of the theoretical orientation of from teacher;
taking, summarizing, generating ques- the study. On the basis of comprehen- 5. Individualization + small group-
tions, outlining, and working in pairs sive reviews that have identified in- statements in the treatment
to summarize sections of materials); structional components that influence description about independent
metacognition (thinking about and student outcomes (e.g., Adams, 1990; practice,individual pacing,
controlling one’s thinking process); Becker & Carnine, 1980; Brophy & individual instruction, and small-
and personal attributions (i.e., child’s Good, 1986; Graham & Harris, 1993; group instruction;
evaluation of the effectiveness of a Pressley & Harris, 1994; Resnick, 1987; 6. Novelty-statements in the
strategy). Rosenshine, 1995; Rosenshine & Stev- treatment description about the
On the basis of these reviews, we ens, 1986; Slavin et al., 1989; Vellutino use of diagrams or picture

categorized studies as reflecting SI in & Scanlon, 1991), we reclustered (or presentations, specialized films or
the following manner. To be classi- reconfigured) the 45 instructional ac- videos, instruction via computers,
fied as an SI model a study must have tivities into approximately 20 clusters specification that a new curricu-
included at least three of the fol- of components for later analysis. We lum was implemented, and an
lowing instructional components: coded the occurrence or nonoccurrence emphasis on new material;
(a) elaborate explanations (i.e., system- of the following instructional compo- 7. Strategy modeling + attribution
atic explanations, elaborations, and / nents : training-statements in the
or plans to direct task performance); treatment description about
(b) modeling from teachers (verbal 1. Sequencing-statements in the processing components or
modeling, questioning, and demonstra- treatment description about multiple steps related to model-
tion) ; (c) reminders to use specific strat- breaking down the task, fading ing from the teacher; simplified
egies procedures; (d) step-by-step
or
prompts or cues, matching the demonstrations modeled by the
prompts or multiprocess instructions; difficulty level of the task to the teacher to solve a problem or
 509

complete a task successfully; statements about short activities, icsinstruction), whereas the Anticipa-
teacher modeling; reminders level of difficulty, necessary tory or preparation response compon-
from teacher to use certain assistance from teacher, simpli- ent characterizes treatment approaches
strategies, steps, and / or proce- fied demonstration by teacher, that activate prior knowledge or pro-
dures ; think-aloud models; and tasks sequenced from easy to vide a precursor to the main instruc-
an enumeration of the benefits difficult, and task analysis; tional activity (e.g., Meichenbaum’s
oftaught strategies; 15. Technology-statements in the [1977] cognitive-behavioral model).
8. Probing / reinforcement- treatment description about The component that reflects the Diffi-
statements in the treatment utilizing formal curricula, newly culty or processing demands of task
description about intermittent or developed pictorial representa- addresses variations in teacher sup-
consistent use of probes, daily tions, specific material or port of the student (e.g., the teacher
feedback, fading of prompts and computers, and of media to
use provides necessary assistance, se-
cues, and / or overt administra- facilitate presentation and quences tasks from easy to difficult;
tion of rewards and reinforcers; feedback; i.e., help is provided to the student
9. Non-teacher-related instruc- 16. Elaboration-statements in the that covaries with the learner’s ability)
tion-statements in the treat- treatment description about and is reflective of activities such as
ment description about home- additional information or mediated scaffolding (e.g., Palincsar
work, modeling from peers, explanation provided about & Brown, 1984). Following explicit sets
parents providing instruction, concepts, procedures or steps, of steps and prompting the use of those
and / or peers presenting or and redundant text or repetition steps (Strategy cues) are important
modeling instruction; within text; activities that underlie SI (Rosenshine,
10. Segmentation-statements in the 17. Modeling of steps by teacher- 1995).
treatment description about statements or activities in the
breaking down the targeted skill treatment descriptions that Control
into smaller units, breaking it involve teacher demonstration Group Parameters
into component parts, and of processes and / or steps the The control conditions were identified
segmenting or synthesizing students are to follow; according to the study’s designation.
components parts; 18. Group instruction-statements However, in a few instances no desig-
11. Advance organizers-statements in the treatment description nation was made within the study, and
in the treatment description about instruction in a small therefore decisions about coding a
about directing children to look group and / or verbal interaction condition as a control treatment were
over material prior to instruc- occurring in a small group com- based on the hypothesis of the study.
tion, directing children to focus posed of students and teacher; If no hypothesis was given, the pre-
on particular information, 19. Supplement to teacher involve- ferred treatment, based on the review
providing prior information ment (but not by peers)- of literature in the primary study, was
about tasks, and stating objec- statements in the treatment assigned to the control condition. The
tives of instruction prior to description about homework coding of the control condition focused
commencing; and parents helping to reinforce on the degree of treatment overlap
12. Directed response / question- instruction; and between the experimental and control
ing-treatment description 20. Strategy cues-statements in the conditions in regard to similarities
related to dialectic or Socratic treatment description about with the experimental group in terms
teaching, teacher directing reminders to use strategies or of teachers, setting, materials, depen-
students to ask questions, multiple steps, teacher verbal- dent measures, exposure to instruc-
teacher and student or students ization of steps or procedures tion (e.g., length of time), primary
engaging in
dialogue, and / or for solving problems, use of focus (e.g., reading, math, etc.), and
teacher asking questions; &dquo;think aloud&dquo; models, and method overlap (number of steps in
13. One-to-one instruction- enumeration of the benefits of treatment vs. control condition that
statements in the treatment strategy use or procedures (see overlap or are the same).
description about activities Note 4).
related to independent practice,
tutoring, and instruction that are These 20 components capture an Effect Size Calculation
individually paced, and instruc- array of intervention approaches. For A weighted Cohen’s d (Hedges &
tion that is individually tailored; example, the Segmentation component Olkin, 1985) was the primary index of
14. Difficulty or processing de- would be a characteristic of analytic ES. Each ES was weighted by the re-
mands of task-treatment and synthesis approaches (e.g., phon- ciprocal in the sampling variance. The
510

dependent measure for the estimate of ES to reflect the primary treatment date; M =
1.97, SD =
3.89 vs. M =
1.60,
of ES was defined as est (d / (1 / v),
=
constructs of each study. SD 2.42). Approximately 15% of the
=

where v is the inverse of the sampling To determine whether a set of ds word-recognition studies and 11% of
variance: shared a common ES (i.e., d was con- the reading comprehension studies
sistent across the studies), a homoge- were published in dissertations;
v =
(Ntrt + Ncrtl)/ (Ntrt X Ncrti) + neity statistic Q, which approximates approximately 18% of the word-
d2/ [2(Ntrt + Nerti)] chi-square distribution with k - 1 de- recognition studies used as their pri-
Cohen’s d is calculated as the differ- grees of freedom, where k is the num- mary outlet the Journal of Learning
between control and ber of effect sizes, was computed Disabilities, whereas the primary out-
ence experimen-
tal treatment posttest mean scores (Hedges & Olkin, 1985). A homoge- let for reading comprehension stud-
neity test (Hedges & Olkin, 1985) was ies was Learning Disability Quarterly
(partialed for the influence of pretest
scores if information is available) di- performed using a Proc GLM SAS (17.9%).
vided by the pooled standard devia- computer program (SAS, 1989) for the In terms of instructional parameters,
tion. To provide a common standard domain categories. When the homo- word-recognition studies were more
deviation across various studies, the geneity of the sample size was rejected likely to have a larger number of in-
pooled (average) standard deviation (i.e., a significant chi-square was structional sessions than reading com-
was calculated from posttest perfor- obtained), the finding was further prehension studies (M = 38.60,
mance of treatment and control con- explored via analysis of the study fea- SD 42.12 vs. M 22.78, SD 23.30
= = =

ditions. When posttest standard devia- tures. A significant chi-square indi- sessions, respectively), but the instruc-
tions could not be calculated (such as cated that the study features signifi- tion time periods were shorter for each
with repeated measures, covariance, cantly moderated the magnitude of session (minutes: M 38.94, SD 29.20
= =

and gain scores), adjustments were effect sizes. When there were two or vs. M 43.14, SD
=
26.45). Sample
=

made in standard deviations. As sug- more levels of a study feature, a Scheff6 sizes tended to be larger in reading-
gested by Hedges and Olkins, outliers post hoc comparison was performed comprehension studies (M 46.90, =

were removed from the analysis of to test for the significant differences SD 65.81) than word-recognition
=

main effects and interactions. Outliers between levels. studies (M 28.23, SD

=
16.23), but =

were defined as effect sizes lying be- there was a comparable distribution
in the number of girls and boys across
yond the first gap of at least 1 stan- Results
dard deviation between adjacent the two types of studies (M 16.71 =

effect-size values in a positive or nega- vs. M = 19.17 for boys and M 8.62 =

tive direction (Bollen, 1989). In exam-

Prototypical Studies vs. SD = 8.12 for girls for word recog-

ining the data, dependent measures Fifty-four studies included measures nition and reading comprehension, re-
from studies that yielded ES > 3.00 of word recognition (159 effect sizes), spectively). The children tended to
were removed from the analysis. For and 58 studies included measures of be slightly older (M 11.62 years,
=

reports of nonsignificant effects unac- reading comprehension (175 effect SD 3.11) in reading comprehension
=

companied by any statistic, we fol- sizes; see Note 5). Twenty studies in- studies than in word recognition stud-
lowed a conservative procedure of cluded both word recognition and ies (M =10.03, SD 1.62). When com-
=

estimating ES at 0.00. reading comprehension measures. As pared to participants in reading com-

There were multiple experimental part of a preliminary analysis, we re- prehension studies, participants in
interventions in some studies we moved the overlapping studies from word recognition studies had slightly
coded (e.g., Lovett, Warren-Chaplin, the analysis; however, the patterns of higher intelligence scores (N 35 =

Ransby, & Borden, 1990). Gleser and the results were comparable and there- M 97.33, SD
=
7.14 vs. N
=
30 =

Olkin (1994) stated that in multiple- fore they were included in the analy- M = 95.74, SD 6.55) and lower read-
=

treatment studies, &dquo;the treatments may ses. ing scores (N = 20 M = 79.97,

all be regarded as instances or aspects A separation of studies by word SD 11.78 vs. N
=
24, M = 85.92,
=

of a common treatment construct&dquo; recognition and reading comprehen- SD 10.78) and math scores (N = 4
=

(p. 351). In addition, they stated, sion indicated that articles were com- M = 88.31, SD 11.46 vs. N = 5
=

&dquo;There is strong reason a priori to parable, respectively, in terms of M 91.58, SD = 12.80). When compar-
=

believe that a composite ES of treat- the year of publication (M =

1987, ing the control condition to the treat-
ment obtained by combing the end SD 6.82 vs. M
= =
1987, SD 5.32);
=
ment condition, both word-recognition
point ESs would adequately summa- number of authors (M=1.33, SD = 1.25 and reading-comprehension studies
rize the effect of treatment&dquo; (p. 351). vs. M 1.31, SD 1.21); and number
= =
were infrequent in their use of the

Thus, although we coded effect sizes of citations from the Social Science same teacher (21.8% vs. 37.0%), mate-

for each individual treatment, we pri- Citation Index (citations were com- rials (19.1% vs. 21.2%), and setting
marily relied on an overall composite puted for 2 years after the publication (29.1% vs. 31%) across conditions.
 511

Internal validity ratings were calcu- internal validity ratings (a rating score sion substantially higher on ex-
were
lated for each study. Internal validity around 7, i.e., a score of 4 reflected perimental (M .81) measures when
=

ratings were based on a scale that takes high internal validity). Overall, ratings compared to standardized measures
into consideration information on of internal validity were higher for (M .45). A chi-square indicated that
=

(a) participant mortality, (b) Haw- word recognition than reading com- no significant differences were found

thorne effects and selection bias, prehension studies. In terms of report- between standardized measures of
(c) comparable exposure between ing reliability of the dependent mea- word recognition (.53) and reading
treatment and control in terms of ma- sures, both types of study provided comprehension (.45), p > .05.
terials, (d) comparable instruction time reliability information on approxi- To determine the independence of
between treatment and controls, (e) com- mately one of three dependent mea- the effect sizes for studies that included
parable (same) teachers between treat- sures. Twenty-three percent of the both reading comprehension and word
ment and controls, (f) procedural va- word-recognition studies and 37% of recognition, the two dependent mea-

lidity checks, (g) practice effects on the reading-comprehension studies sures were correlated and found to be
the dependent measure, (h) floor and reported some measure of treatment significant, r(18) = .49, p < .05. Thus,
ceiling effects, (i) interrater reliability, fidelity (integrity). a high degree of interdependence

(j) regression to the mean ruled out as emerged among measures, suggesting
an alternative explanation of findings, that subsequent analyses must (a) sep-
Domains of Instruction
and (k) homogeneity of variance be- arate reading comprehension from
tween groups. An abbreviated form Table 1 shows the unweighted and word recognition studies and/or
(Items 2, 4, 8) and a complete form
1, weighted mean Cohen’s d effect (b)control for the intercorrelation
(Items 1-11) wereused to rate stud- sizes for the general categories and among measures within studies (as
ies. For each of the 11 items, a score of subcategories. The word-recognition done in a regression analysis). ,

1 reflected internal validity control, and comprehension categories were

2 reflected no validity control, and further divided by standardized
Main Effects
3 reflected no validity information (norm-referenced) and experimental
(scores ranged from 11 to 33). The (researcher-developed) measures. Tables 2 and 3 show the WLS mean
mean scores for the full and abbrevi- Based on the weighted effect sizes re- effect sizes as a function of method-
ated scales on internal validity, respec- ported in Table 1, those areas that ap- ology, article characteristics, sampling
tively, were 22.56 (SD 2.43) and 7.63
=
proached Cohen’s (1988) threshold of characteristics, and independent vari-
(SD = 1.79) for word recognition stud- .80 for a &dquo;large effect&dquo; were measures ables for studies that included word-
ies, and 21.77 (SD 2.67) and 7.14
=
of reading comprehension (.72). When recognition and reading compre-
(SD 1.76)
= for reading comprehen- the categories were divided into sub- hension measures, respectively. The
sion studies. Most studies on the ab- categories, however, the higher effect number of independent samples was
breviated form yield low to moderate sizes related to reading comprehen- also computed (see Cooper, Nye,

TABLE 1
Weighted Mean Effect Sizes for Word Recognition and Reading Comprehension Measures

Note. From Interventions for Students with Learning Disabilities: A Meta-Analysis of Treatment Outcomes (pp. 146-159), by H. L. Swanson, with M. Hoskyn
and C. Lee, 1999, New York: Guilford. Copyright 1999 by Guilford. Adapted with permission. n = number of studies; K = number of dependent measures.
**p < .001.
 512

TABLE 2
Word Recognition Effect Size Estimates Averaged Within Studies

(table continues)
 513

(Table 2 continued)

Note. N = 54. From Interventions for Students with Learning Disabilities: A Meta-Analysis of Treatment Outcomes (pp. 146-159), by H. L. Swanson with
M. Hoskyn and C. Lee, 1999, New York: Guilford. Copyright 1999 by Guilford. Adapted with permission.
apartialed for the methodology composite score.
 514

TABLE 3
Reading Comprehension Effect Size Estimates Averaged Within Studies
 515

Note. N = 58. From Interventions for Students with Learning Disabilities: A Meta-Analysis of Treatment Outcomes (pp. 146-159), by H. L. Swanson with
M. Hoskyn and C. Lee, 1999, New York: Guilford. Copyright 1999 by Guilford. Adapted with permission.
apartialed for the methodology composite score. bin some cases, total K # 58 because level of the variable cannot be separated out. CWLS means with
methods partialed.
*p < .05. **p < .01.
516

Charlton, Lindsey, & Greathouse, 1996, larger effects sizes for small (< 25) and ite score was used in the subsequent
for the complete rationale). When the large (> 100) samples when compared analysis to qualifythe outcomes re-
number of independent samples ex- to moderate (> 24 and < 101) sample lated to ES estimates. Studies that in-
ceeded 2 x K (the number of studies), conditions. In terms of standardized cluded the following were assigned a
then the studies included than
more dependent measures, reading compre- score: (a) instructional sessions greater

one experimental treatment. When the hension studiesyielded significantly than 10 (selection of this variable was
number of independent samples was lower effect sizes on standardized based on the assumption that the in-
lower than 2 x K, then there was reli- measures when compared to experi- tensity of instruction-as reflected in
ance on some studies that used the mental measures, whereas no sig- the number of sessions-yielded more
same participants in multiple treat- nificant difference emerged between reliable and stable outcomes than
ments. In cases of repeated measures, such measures for reading recognition shorter intervention sessions); (b) ran-
effect sizes were recalculated via a studies. dom assignment of treatment; (c) mea-
formula that took into consideration Both types of studies were sensitive sures of treatment integrity; (d) utili-

the correlation between the control and to teacher effects. Studies that used a zation of standardized tests; (e) high
treatment conditions. A chi-square different teacher between treatments ratings (score of 4) on the internal
based on the WLS analysis was pre- yielded higher effect sizes than stud- validity abbreviated scale (that num-
sented for all comparison measures. ies that controlled for teacher effects. ber reflects highest possible rating on
When the degrees of freedom were For reading comprehension studies, items on the variables related to the
greater than 1, a Scheff6 test was used those that did not specify how mate- comparability of setting, materials, and
to make comparisons between the es- rials between control and treatment teachers across conditions and the re-
timates of the mean ES. conditions differed yielded lower ef- porting of procedural validity); and
As shown in Tables 2 and 3, for fect sizes than those that did. In addi- (f) high control and treatment condi-
word-recognition and reading compre- tion, reading comprehension studies tion overlap in terms of steps and pro-
hension studies, no significant differ- that did not report where the treat- cedures (at least three instructional
ences were found among the estimates ment occurred yielded larger effect steps and / or procedures overlapped).
of ES as a function of the number of sizes that those that did. No signifi- The amount of psychometric data re-
treatment sessions, internal validity cant differences in ES were found ported on the sample was also in-
ratings, and measures of treatment among reading recognition studies as cluded in the methodological compos-
integrity (measures of the degree to a function of setting. ite score (if additional psychometric
which treatment was implemented as Both word recognition and reading information beyond an IQ score was

planned).No significant differences comprehension studies were suscep- reported-e.g., reading scores-the
in ES were found among the types tible to variations in the control con- study was weighted positively). For
of random assignment for word- ditions. For word recognition studies, each of the seven variables weighted
recognition studies, whereas reading the lowest effect sizes occurred when positively and negatively, a score of 2
comprehension studies that indicated the focus of the control condition was or 0 was assigned for each variable,

no random assignment of treatment &dquo;on skills other than reading&dquo; when respectively. For each study, the
yielded larger effect sizes than other compared to studies that included weighted score varied from 14 to 0,
conditions. No significant differences control conditions that focused on the with 14 reflecting methodologically su-
in ES found as a function of the
were same domain (i.e., reading) or when perior studies. The mean methodologi-
degree overlap in instructional pro-
of studies reported no information on the cal composite score for word recogni-
cedures between the control and treat- focus of the control group. For read- tion studies (M =
5.77, SD =
2.81)
ment conditions for word-recognition ing comprehension studies, no differ- was higher than the
composite score

studies, whereas reading comprehen- ences were found among effect sizes for reading comprehension studies
sion studies that reported very little related to variations in the focus of (M 4.82, SD 2.90).
= =

overlap in instructional procedures be- control situations, but studies that As shown in Table 2, variations in
tween treatment and control condi- failed to report the focus of the con- the methodological sophistication of
tions yielded higher effect sizes than trol condition yielded the lowest ef- word recognition studies had mini-
studies with high overlap. fect sizes when compared to com- mal influence on the magnitude of ES,
Both reading recognition and read- peting conditions. whereas Table 3 shows that reading
ing comprehension studies were sen- Because several methodological vari- comprehension studies high in meth-
sitive to sample size: Generally, studies ables were clearly related to the mag- odological sophistication yielded sig-
with small sample sizes yielded larger nitude of ES, it was necessary to create nificantly lower effect sizes than
estimates of ES than those with larger a composite score related to internal studies of low sophistication. Approxi-
samples. Reading comprehension validity and methodological sophisti- mately 82% of the reading compre-
studies yield a bow-shaped effect, with cation. This methodologically compos- hension studies and 74% of the word
 517

recognition studies yielded low meth- report such information. Studies that mum number of components for an
odological ratings. reported no psychometric information SI model classification and those be-
In terms of article characteristics, noon intelligence yielded significantly low the minimum number, as well as
significant differences in ES were higher effect sizes than studies that between those studies at or above the
found for word recognition studies as reported average or low-average in- minimum threshold to be classified
a function of funding, publication telligence scores. In addition, studies as a DI model and those below the

outlet, or type of setting used for in- that reported no scores on reading minimum number. When SI and DI
tervention. For reading comprehension severity yielded higher effect sizes were divided into four general mod-

studies, those that report a funding than studies that reported reading se- els, the Combined Model yielded
source yielded lower effect sizes than verity scores. larger effect sizes than DI alone, SI
studies that did not report a funding As can be seen in Tables 2 and 3, alone, and non-SI & non-DI models
source. Reading comprehension stud- few studies provided information that for studies that included measures of
ies that were published in disserta- allowed us to investigate the effects reading comprehension. No significant
tions or technical reports yielded of gender on treatment outcomes. (ps > .05) differences in ES were found
significantly smaller effect sizes than Thus, we calculated the median ratio between the competing models (DI
studies that were published in jour- of studies that reported the number alone vs. SI alone vs. Non-SI and Non-
nals. In terms of setting, reading com- of boys and girls in each study. We DI models). Although a trend was
prehension studies that reported the computed the gender ratio (number found for the direct-instruction-only
setting in which the treatment occurred of boys / total in the sample) and di- model to yield higher effect sizes than
had higher effect sizes than those that vided the studies by the median ratio the competing models for word rec-
did. As a follow-up to this analysis, (.70) across the 94 studies. Those stud- ognition studies (LSM .70), no sig-
=

those studies that reported implement- ies that reported the number of boys nificant differences were found be-
ing treatment in the general educa- and girls in the sample and included tween models. This finding is qualified
tion classroom and resource room had more than 69% boys in their sample in the subsequent analysis.
higher effect sizes than those studies were compared to studies that re-
reporting that treatment occurred in a ported less than 70% boys in the
self-contained (clinic) setting, or stud- sample. These two categories were Exploratory Analysis
ies that failed to report the setting of compared with studies in which the The next set of analyses identifies in-
treatment. The finding on general edu- gender ratio of the sample could not structional components that played a
cation classrooms should be inter- be computed. As shown in Tables 2 major role in predicting ES estimates
preted cautiously because only one and 3, only 5 word recognition and regardless of the instructional (e.g., DI
study directly stated they imple- 11 reading comprehension studies only) model. We identified the instruc-
mented training in such a classroom. were considered to have a low-boy tional components that were most
Thus, some of the &dquo;no information ratio. For word recognition studies (see important in predicting ES, as well as
studies&dquo; may have included treatments Table 2), gender ratios did not influ- those components that, when com-
in the general education classroom. ence the magnitude of ES. Table 3 bined, accounted for most of the vari-
In terms of sample characteristics, shows that larger effect sizes occurred ance in treatment outcomes. To deter-
no significant differences in ES were for reading comprehension studies mine the relationship between the
found between studies that reported where gender ratios could not be cal- 20 instructional components and ES
samples who were at least 1 year be- culated compared to studies where estimates, WLS fixed-effects regression
hind their peers in achievement and gender ratio could be calculated. An- models were fitted to the empirical
studies reporting that the participants other sample variable considered was values, following methods recom-
were 1 year or less behind their peers. the age of the sample; in contrast to mended by Harwell (1992) and de-
For word recognition studies, no sig- reading comprehension studies, how- scribed by Hedges and Olkin (1985;
nificant differences in ES were found ever, word recognition studies that see also Hedges, 1994a, 1994b). For

as a function of the psychometric char- included younger students yielded the WLS regression analysis, all in-
acteristics of the sample. In contrast, larger effect sizes than those that in- structional components were dummy
several psychometric characteristics cluded older students. coded (coded 1 for occurrence of com-
distinguished reading comprehension The final analysis of main effects ponents and 0 for nonoccurrence) for
studies in terms of the magnitude of considered general models of instruc- the subsequent analysis. To facilitate
ES. Studies that reported more psy- tional intervention. The weighted mean the interpretation of the components
chometric information (i.e., intelli- ES partialed out the influence of the for the reader, the component num-
gence, reading, and mathematics methodological composite score. Ta- ber is provided beside the descrip-
scores) yielded significantly smaller bles 2 and 3 showed the comparisons tion in the Method section. The com-
effect sizes than studies that did not between studies at or above the mini- ponent number is described in the
518

Method section under the subhead- the &dquo;best model&dquo; option. The adjusted practice, (e) independent practice, and
ing of treatment variables
(e.g., Com- R2 procedure was used to find the (f) formative evaluations. The compo-
ponent 1 sequencing, Component
=
model with the highest R2 within the nents in our synthesis that matched
10 segmentation).
=
ranges of ES. The methodological com- these practices were Orientation to a
Three approaches were used to con- posite and chronological age
score Task (Component 3; see the Method
trol for preselection biases (control for variables entered first into the
were section), Drill-Repetition-Practice
order) when entering the instructional equation. For word recognition stud- (Component 2), Teacher Modeling of
components, and for capitalization on ies only, the adjusted R2 of .02 for Steps (Component 17), Sequencing
chance in the WLS regression analy- methodology and age increased to an (Component 1), and Systematic Prob-
sis. First, we used a stepwise selection adjusted R2 of .19 when the compo- ing (Component 8). We refer to these
procedure in which the order of entry nents of treatment included Sequenc- five components as an instructional core.
was determined via a mathematical ing (1), Segmentation (10), and Ad- To be precise, we wanted to deter-
maximization procedure. That is, af- vanced Organizers (11). These three mine if the amount of variance in ES
ter the methodological composite scores components contributed 17% of the estimates for (a) age and methods
were entered, the component with the variance in predicting ES estimates. alone; (b) age, methods, and the in-
largest correlation to ES entered, fol- For reading comprehension studies structional core alone; and (c) age
lowed by the second component with only, the adjusted R2 when entering methods, instructional core, setting
the largest semipartial coefficient cor- methodology and age was .08 but in- plus components of theoretical inter-
relation, and so on. Chronological age creased to .30 when components re- est, varied significantly. The setting
of the sample was also left to covary. lated to Directed Response and Ques- components included small group,
For word recognition studies, the re- tion (12), Elaboration (16), Modeling adaptive (individualized), or one-to-
sults indicated that the Anticipatory of Teacher Steps (17), Small Group one tutored instruction. The variables
or Preparation component (Compo- Instruction (18), and Strategy Cues (20) of theoretical interest included seg-
nent 3) contributed approximately 6% entered into the equation. These five mentation, scaffolding (control task
of the variance in predicting ES, fol- components contributed 22% or the difficulty), and cuing the use of strat-
lowed by segmentation (Component variance to ES estimates. egies. These components were selected
10), R2 increment of .04. No other on the basis of the reviews of litera-

variables were significant, ps > .05. ture on DI and SI and also emerged in
Hierarchical Regressions
The complete model was significant, our exploratory analysis as important
R2=.10, X2(2, N = 54) 15.77, p < .001.
=
The final set of analyses consisted of a variables.
For reading comprehension studies, series of hierarchical regressions us- Three models were tested beyond
the results indicated that the method- ing a priori selected instructional com- the instructional core. One model fo-
ological composite score contributed ponents to predict ES estimates. (The cuses on intense individual instruction

approximately 8% of the variance in intercorrelations for the analysis can and segmentation, which is suggested
predicting effect sizes, R2 .076, be obtained from the author.) We spe-
=
as necessary for phonics instruction (Tor-

x2(1, N 58) 23.46, p < .001. The first cifically tested whether the type of
= =
gesen et al., 1997; Vellutino et al., 1997).
instructional component to enter the instructional setting (intense one-to- The second model is characteristic of
regression model was modeling of one, or small interactive groups) and several reciprocal teaching models that
teacher steps (Component 17), incre- three components identified frequently include small, interactive group instruc-
‘ment .06; followed by Directed
=
in the literature as critical to reading tion and control of task difficulty (scaf-
Response/Questioning (12), increment = instruction (Segmentation, Strategy folding ; Palincsar & Brown, 1984). The
.06; followed by age, increment .05; Cuing) contributed significant variance third model focuses on small-group
followed by technology (15), increment to estimates of ES after sound instruc- strategy instruction (e.g., Marks et al.,
=
.05. The complete model accounted tional practices were taken into con- 1993; Pressley et al., 1992). Compet-
for 30% of the variance, R2 .298, x2(5, sideration. That is, we determined ing models are hybrids of these three
=

N 58) 14.72, p < .05.

= =
whether the competing models con- approaches.
Second, we identified those compo- tributed beyond a common effective Table 4 presents the results of the
nents that, when combined, yielded instructional core suggested by previ- WLS hierarchical regression analysis.
the highest adjusted R2 in predicting ous reviews (see Rosenshine, 1995; As shown, the methodological and age
ES estimates. The squared multiple Rosenshine & Stevens, 1986; Slavin, variables were always entered first into
correlation, adjusted for the number Stevens, & Madden, 1988). These re- the equation, followed by the instruc-
of explanatory variables, was used to views have suggested that sound in- tional core. Variations in the instructional
quantify the predictive power of the structional practices include (a) daily setting (individuals; small, interactive
given model. The SAS regression pro- reviews, (b) statements of an instruc- group) and then the components of seg-
gram (SAS Institute, 1989) was used tionalobjective, (c) teacher presenta- mentation, controlling task difficulty,
to conduct the statistical analysis, via tions of new material, (d) guided and strategy cuing were entered into
 519

the equation after the methodological

TABLE 4
Weighted Least Square Hierarchical Regression for Word Recognition and composite score, age, and the instruc-
tional core. Four findings in our analy-
Reading Comprehension Studies
sis were important. First, for word rec-
ognition, segmentation did not enter
significantly into the model after the
basic core model was entered into the
equation. Thus, it appears that a clear
orientation to task, drill-repetition-
practice, sequencing, teacher modeling,
and systematic probing may supersede
effects related to segmentation. How-
ever, we are cautious in this comment
because the intercorrelation between
segmentation and sequencing was
high (phi correlation .73), suggest-
=

ing that both these components are

usually part of the same instructional
program. Regardless, it seems that
when careful sequencing of informa-
tion occurred (i.e., step-by-step pre-
sentation), the effects of segmentation
(broadly defined-i.e., word segmen-
tation to text segmentation) were more
difficult to detect. Second, the scaf-
folding component (control of task
difficulty by adapting materials and
cuing) did not contribute significant
independent variance to reading com-
prehension after the instructional core
was entered into the model. Third,
combining group and individual in-
struction into a total model did not
improve the R2 in predicting word
recognition and reading comprehen-
sion. Interactive instruction within a
small group made its most important
contribution to measures of reading
comprehension, whereas neither the
one-to-one the small-group in-
nor
struction variable contributed signifi-
cant variance to ES estimates for the
word recognition studies. Finally, both
small group interactive instruction and
strategy cuing contributed significant
variance to reading comprehension
beyond the core model.

Influence of Sample Definition

onDomain and Treatment
Outcome
The final analysis determined whether
various configurations of IQ and read-
(table continues) ing scores influenced treatment out-
520

(Table 4 continued) culated were .25, .61, .41, .50 for com-
bination, DI-only, SI-only, and Non-
DI and Non-SI studies, respectively.
A Scheff6 test (alpha =
.05) indicated
that the studies that used a DI-only
model and included samples that met
cutoff score criteria yielded higher ef-
fect sizes than studies that included
competing models (DI only > Combi-
nation > SI only Non-SI and Non-
=

DI). Effect sizes for the DI-only model

were significantly (p < .05) higher
when compared to studies that in-
cluded the combination model when
cutoff scores could not be computed
anumber refers to order of entry into the regression model. (DI only > Non-DI and Non-SI SI =

*p < .05. **p < .001. only > combined).

For reading comprehension studies,
the WLS analysis, with the influence
comes. Four contrast variables related below the 25th percentile and those of the methodology score partialed out,
to the definition of LD were created. that reported a score above the 25th indicated that significant effects oc-
The influence of each of these con- percentile. Each contrast variable and curred for high IQ discrepancy defined
trast variables on the magnitude of the four treatment conditions were samples (mean IQ 97.77; SD 2.60;
= =

ES was considered across the four submitted to a WLS regression analy- mean reading score 79.83, SD 6.68)
= =

treatment models (combination, DI sis, with the methodological score par- vs. studies in which discrepancies were

only, SI only, and low SI and DI thresh- tialed out, as a function of word rec- in the low IQ range (mean IQ 90.28; =

old models) in a WLS regression analy- ognition and comprehension studies. SD 2.13; mean reading score 81.32,
= =

sis. Only the significant results are re- SD 8.22) or could not be computed
=

The first contrast variable (defined ported below. x2(2, N 58) 6.54, p < .05. (Note the
= =

as meeting cutoff score criteria vs. For word recognition studies, a WLS large standard deviation for reading,
other, or not meeting cutoff score cri- analysis indicated that Contrast 1 (cut- which accounted for the fact that mean
teria) compared studies that met a off criterion definition) yielded signifi- scores for IQ were in the same range
cutoff reading at or below the
score in cant outcomes. A follow-up analysis as mean reading scores.) The results
25th percentile (the 25th percentile is compared studies, via a 2 (cutoff score indicated that studies in which dis-
a standard score of 90) and reported criteria vs. non-cutoff score criteria) x crepancies between IQ and reading
intelligence scores at or above 85 to 4 (treatment, i.e., combination, DI could not be computed (N 42, WLS =

those studies in which IQ and / or read- only, SI only, Non-DI and Non-SI mod- M = .84) yielded larger effect sizes than
ing scores were not reported or failed els) design, with the methodological those where group discrepancies could
to meet criteria. A second contrast composite score partialed from the be computed. However, a Scheff6 test
variable (referred to as high IQ vs. low analysis. Studies that met the cutoff indicated that for studies in which
IQ discrepancy defined samples) com- score criteria yielded higher ES scores discrepancies could be computed, ef-
pared studies that included samples (N = 15, LSM = .78; mean IQ = 98.21, fect sizes were significantly (p < .05)
differing by 15 or more standard score SD = 8.95; mean reading = 76.31, larger (N 8, WLS M .67) for stud-
= =

points between reported IQ and read- SD = 8.26) than those that did not ies that reported IQs in the low-
ing, but whose IQ was above 90 or (N = 39, LSM .47), x2(1, N 54) =
= =
average range (< 91) than those that
below 91. A third contrast (referred to 5.77, p < .05. A significant interaction reported IQs in the high-average range
as high vs. low discrepancy) compared emerged for the definition x treatment (IQs > 90; N =
8, WLS M =
.49). No
studies that had discrepancies > interaction, x2(3, N 54) 7.39, p < .05.
= =
significant interaction emerged related
15 points between reported sample IQ Weighted least square means for stud- to type of treatment. The significant
and reading means to those studies ies that met cutoff score criteria were main effect for treatment is reported
with low discrepancy differences (dif- .92, 1.06, .55, .56 for combination, DI- in Table 3 (last section).
ference was < 15). A final contrast (re- only, SI-only, and Non-DI and Non- In summary, the contrast variables
ferred to as reading-level only) compared SI studies, respectively. Weighted least that did not yield a significant rela-
the magnitude of ES as a function of square means for studies in which tionship (ps > .05) to the magnitude of
studies that reported a reading score cutoff score criteria could not be cal- effect size were (a) those that com-
 521

pared the degree of discrepancy (stud- published reports on effective inter- control and experimental condi-
ies with mean scores in IQ and read- vention with children with dyslexia tions yielded significantly higher
ing > 15 and those < 15) and (b) those (e.g., Foorman et al., 1998; Vellutino effect sizes than studies that had
that reported reading scores < 91 (25th & Scanlon, 1991), suggesting that there a high overlap in terms of

percentile). Contrast variables that mod- is some collective evidence that per- instructional components (se-
erated effect sizes for reading recog- sistent word recognition and compre- quences, steps, methods, and
nition were not the same as those that hension deficits are amendable to procedures).
moderated the effect sizes from read- remedial training. There are several 5. Studies that did not report the
ing comprehension. Studies meeting additional findings of interest. setting in which the intervention
cutoff score criteria yielded higher ef- First, the results showed that out- occurred yielded significantly
fect sizes than studies not meeting comes were influenced by variations lower effect sizes than those that
cutoff score criteria for word recogni- in methodological variables. Word did report the setting.
tion studies. These findings were quali- recognition measures were less sus-
fied, however, by treatment for the ceptible to methodological variation In sum, artifacts related to method-
word recognition studies. For reading than reading comprehension; how- ology have a profound influence on
comprehension, studies that reported ever, studies in both word recogni- treatment outcomes, especially in the
differences (mean sample difference tion and reading comprehension were domain of reading comprehension.
in each study at or greater than 15 stan- influenced by confounds related to However, we also found that when
dard score points) between mean IQs sample size (smaller sample sizes we partialed out the weighted com-
and mean reading scores yielded low yielded larger effect sizes than larger posite score related to methodology,
effect sizes when IQs were in the high- samples) and teacher (different teach- significant effects related to the type
average range, when compared to ersyielded higher effect sizes than of treatment emerged.
other conditions. when teacher effects were controlled). Second, only a few instructional
Additional variables related to word components moderated treatment out-
recognition that influenced effect size comes. Interestingly, all four treatment

General Discussion were age (younger sample yielded models (combined, DI alone, SI alone,
larger effect size than older sample) and models that failed to meet strat-
This article synthesizes intervention and the type of control group (para- egy and DI component thresholds)
research for students with LD that doxically, control conditions that did yielded WLS mean estimates > .40 on
includes word recognition and read- not focus on reading yielded lower both word recognition and reading
ing comprehension measures. The re- effect sizes than those that did). Sev- comprehension measures. These find-
sults from the meta-analysis were eral isolated methodological variables ings were problematic for interpreta-
consistent with the literature indicat- that yielded significant effects on the tion, but were characteristic of most
ing that a combined instructional magnitude of ES estimates for reading syntheses of intervention research (a
model that includes components of comprehension studies are summa- bias toward positive outcomes; Lipsey
both strategy and direct instruction rized as follows: & Wilson, 1993; also see Weisz et al.’s,
positively influences reading compre- 1995, discussion of the &dquo;dodo&dquo; verdict).
hension performance (e.g., Graves & 1. Studies in which control and One means to circumvent this prob-
Levin, 1989; Harris & Pressley, 1991; treatments occurred in a different lem was to isolate the instructional
Johnson et al., 1997) and that direct setting (i.e., classroom and activities that were shared among ef-
instruction improves word recognition school) yielded significantly fective treatments. Thus, our con-
(e.g., Foorman, Francis, Fletcher, higher effect sizes than those in clusions about certain instructional
Schatschneider, & Mehta, 1998; Lovett the same classroom setting. models (combined vs. DI only) were
et al., 1994). Specifically, the synthesis 2. Studies that relied on experimen- qualified because we assumed that
indicated that when compared to other tal measures yielded significantly studies that included effective treat-
models, the combined model yielded higher effect sizes than studies ments (i.e., yielded high effect sizes
ES estimates (M 1.15) that exceeded
=
that relied on standardized relative to the control condition) but
Cohen .80 criteria for a substantive measures. reflected diverse theoretical perspec-
finding in the domain of reading com- 3. Studiespublished in dissertations tives, shared some commonalities re-
prehension, and the DI approach ex- yielded significantly lower effect lated to instructional components,
ceeded this criteria when studies that sizes than those published in tactics, and/or procedures. For ex-
included word recognition measures journals. ample, there are some similarities as
met cutoff score criteria (M 1.06).
=
4. Studies that had
a low overlap in well as clear differences between the
The magnitude of these findings was terms of the number of instruc- DI and combination models. For ex-

consistent with a few syntheses of tional components between ample, two components frequently
522

shared between the combined model 1. Directed responselquestioning: breaking down the task, fading of
and the DI-alone model are sequenc- treatment description related to prompts or cues, matching the
ing and segmentation. Both models dialectic or Socratic teaching, the difficulty level of the task to the
also share the common format of teacher directing students to ask student, sequencing short activi-
small-group and individual instruc- questions, the teacher and ties, and / or using step-by-step
tion. The combined model is more students or students engaging in prompts.
likely, however, to emphasize novelty dialogue, and / or the teacher 2. Segmentation: statements in the
(e.g., new materials), attributions, and asking questions. treatment description about
teacher modeling of processing steps 2. Control difficulty of processing breaking down the targeted skill
to perform a task when compared to demands of task: treatment state- into smaller units, breaking it into
the other models. ments about short activities, level component parts, segmenting
To further identify the instructional of difficulty controlled, teacher and / or synthesizing component
and methodological ingredients that providing necessary assistance, parts.
make up an effective intervention, this teacher providing simplified 3. Advanced organizers: statements in
synthesis did not rely on the label of demonstration, tasks sequenced the treatment description about
the independent variable provided by from easy to difficult, and / or task directing children to look over
the authors to characterize treatments. analysis. material prior to instruction,
Instead, the independent variables 3. Elaboration: a treatment statement children directed to focus on
were coded on the reported methods about additional information or particular information, providing
and instructional components that explanation provided about prior information about task,
made up the treatments. Based upon concepts, procedures, steps, and / or the teacher stating
the reported instructional component and / or redundant text or repeti- objectives of instruction prior to
information, studies sorted into
were tion within text. commencing.
distinct and overlapping strategy and 4. Modeling by the teacher of steps:
DI models (i.e., combined, SI alone, statements or activities in the The importance of these findings is
DI alone, and Non-DI and Non-SI or treatment descriptions that thatonly fewa components from a
low-threshold models). We did not di- involve modeling by the teacher broad array of activities enhance treat-
rectly compare (although the data are in terms of demonstration of ment outcomes.
there), for example, ES differences in processes and / or steps the Third, substantial changes emerged on
our synthesis between peer-mediated students are to follow to solve both word recognition and comprehension
instruction (e.g., Mathes & Fuchs, 1993), the problem. measures. When estimates were col-

computer speech support (e.g., Lund- 5. Group instruction: statements in lapsed across treatment conditions,
berg & Olofsson, 1993), vocabulary in- the treatment description about effect sizes within a 95% confidence
struction (Pany, Jenkins, & Schreck, instruction in a small group, interval for both word recognition and
1982), cognitive learning strategies and / or verbal interaction occur- comprehension ranged from .52 to .77.
(e.g., Williams et al., 1994; Wong & ring in a small group with The majority of studies in the domain
Jones, 1982), analytic phonics (Lovett students and / or teacher. of reading comprehension that used
et al., 1994), synthetic phonics (e.g., 6. Strategy cues: statements in the experimental measures met Cohen’s
Foorman et al., 1997), whole language treatment description about criterion of .80 as a substantive finding.
(e.g., Lerner, 1978), reciprocal teach- reminders to use strategies or However, the magnitudes of ES were
ing (e.g., Klingner & Vaughn, 1996), multiple steps, the teacher significantly inflated on experimental
and so forth. Instead, we focused on verbalizing steps or procedures to measures (.81) when compared to stan-
the overlapping components (those solve problems, use of ’think dardized measures (.45). Although the
shared by all effect treatments) and aloud models,’ and / or teacher magnitude of effect sizes related to
nonoverlapping instructional compo- presenting the benefits of strategy word recognition were smaller, they
nents that predicted treatment out- use or procedure. were stable across experimental and
comes. We found that regardless of standardized measures (.53 vs..62).
the general model of instruction, only The important instructional compo- We expected that changes in read-
a few instructional components in- nents for word recognition that emerged ing measures, especially for word rec-
creased the predictive power of treat- from our regression analysis were the ognition, would be in the low range.
ment effectiveness beyond what could following: The persistence of these word recog-
be predicted by variations in method- nition and comprehension deficits
ology and age. For reading comprehen- 1. Sequencing: statements in the commonly attributed to children with
sion, the components were as follows: treatment description about LD would suggest that substantial
 523

changes in ES depend systematically the results clearly showed that IQ mod- segmentation and were not examined
on the passage of time, where time in erated treatment outcome. separately. Thus, it is possible that
this case is indexed by the number of Fifth, an instructional core of compo- different outcomes may have emerged
intervention sessions. Such was not nents that included careful sequencing of if our coding of segmentation had been
the case in this synthesis: Across 92 activities and materials, well as sys-
as further refined. However, it is impor-
studies, the weighted effect sizes were tematic feedback, partialled the influence tant for the reader to note that seg-
in the moderate to high range, and of segmentation on word recognition. The mentation is seldom used as a single
the magnitude of these effect sizes did results from the exploratory analysis manipulated variable in a treatment
not vary significantly with short or and the components that made up the sequence. In most cases, treatment
extended numbers of intervention ses- combination model and DI-only model comparisons that include segmenta-
sions (see Tables 2 and 3). Thus, the clearly indicated that the &dquo;segmenta- tion are confounded with other com-
magnitude of the effect sizes for word tion&dquo; component positively influenced ponents, as well as variations in control
recognition and comprehension was the magnitude of treatment outcomes. conditions and sample size. To illus-
stable over time (i.e., the number of The hierarchical regression illumi- trate, some individual studies in our
treatment sessions). nated, however, why equivocal find- synthesis that included segmentation
Fourth, cutoff scores qualified treatment ings related to the influence of segmen- will be briefly reviewed.
effects in word recognition studies, and tation on transfer measures, such as In a study by Kennedy and Backman
variations in IQ-reading discrepancy real word recognition, might have (1993), participants were instructed
scoresqualified treatment effects for read- emerged. Our synthesis research sug- in a Lindamood Program, which empha-
ing comprehension studies. The impor- gests that perhaps the careful sequenc- sized identification of speech sounds
tant finding for word recognition was ing of materials and information across and orthographic symbols. The Linda-
that studies that allowed for the cal- experiments may have partialled out mood Program was a supplement to
culation of cutoff scores (reporting the effects that segmenting informa- regular private school LD instruction,
standardized IQ scores at 85 and above tion (such as in a phonics programs) which included phonics. Control par-
and reading scores below 90 [25th had on word recognition. Although ticipants also received private instruc-
percentile]) yielded higher effect sizes there has been some debate about tion that included phonics. With only
than the remaining studies. In addi- whether the emphasis of intervention 10 participants in each condition, the
tion, it seems that studies that allowed in sight word reading should be di- ES on the Slosson Oral Reading Test
for the calculation of cutoff scores rected at the whole word level, and / was .14 and the Gray Oral Reading

yielded higher effect sizes for the DI- or at the analytic level (isolated Test was .28. As in other studies, high
only model when compared to other sounds), and / or at the synthesis level effect sizes were found for the pro-
instructional models. Interestingly, (sound units), the present synthesis cesses directly related to instruction
both variables (IQ and reading) in iso- suggested that this debate might be (phonics), but small effect sizes were
lation did not significantly influence better addressed, at least at the sight found related to real-word recognition.
treatment outcomes (see Table 2). Al- word level, when systematic compari- In another study, Brown and Felton
though this finding was expected for sons are made between treatments that (1990) found that a structured phon-
IQ, it was assumed that severity of involve segmentation with an instruc- ics approach (using the Lippincott
reading deficits would moderate treat- tional core that includes sequencing, Basic Reading Program) produced
ment outcomes. There was a trend that feedback, and daily review. more gains than literature-based or
readers with more severe problems Because the above finding related whole word instruction for first grad-
yielded higher effect sizes (M .74) =
to the hierarchical regression model ers at risk for reading deficits (i.e.,
than other categories of readers, but was not predicted, further comment identified as having poor phonologi-
these findings were not significant. In is necessary. Although half of the in- cal processing). Although the ES after
contrast to word recognition studies, terventions in this synthesis that used 2 years of instruction was .52 on the
effect sizes for reading comprehension word recognition measures used some Word Identification subtest of the Wood-
measures were moderated by high and type of word segmentation (55%) in cock Reading Mastery Test (WRMT)
low IQ scores. Studies of individuals the presentation of treatment, these (in favor of the phonics condition), it
with low-average IQs with at least a studies clearly varied in their level of is important to note that two different
15-point discrepancy yielded higher phoneme segmentation (e.g., letter settings were used for comparisons
effect sizes than studies of persons sounds, intrasyllabic units of onset- (which means by inference that at least
with a discrepancy range but IQs rimes to sensitivity [phonological two different teachers were used). As
above 90. Although there was a clear awareness] of letters and phonemes). we have indicated in our Results sec-
bias for studies that do not report IQ All these variations were coded in the tion (and see Table 2), whenever dif-
and reading information (see Table 3), present synthesis as an example of ferent settings or teachers are used
524

when comparing treatment to controls, cient, condition for learning to read a measure of scaffolding, was signifi-
effect sizes are highly inflated. (e.g., Adams, 1990; Vellutino & Scan- cant when left
to compete with other
Some studies include control condi- lon, 1991 ). components but did not contribute to
tions that are just as effective as the Finally, strategy cuing and small-group, instruction when instruction included
treatment conditions. For example, interactive instruction contribute signifi- control of sequencing. Unfortunately,
Torgesen, Wagner, Rashotte, Alex- cant variance to estimates of reading com- many of the studies that included scaf-
ander, and Conway (1997) suggested prehension beyond the instructional core folding yielded low methodological
gains for students with reading dis- model. Some of the more well-designed composite scores, and therefore this
abilities (approximately 10 years old) studies on strategy cuing and small- component’s independent contribution
who received one-to-one tutoring and group instruction were provided by may be difficult to evaluate.
synthetic and analytic phonics instruc- Graves (1985) and Wong and Jones
tion with a skilled clinician for 2 hours (1982). Graves reported successful out-
daily over 8 weeks. However, when comes with students with LD related Qualifications
we calculated effect sizes for real-word to comprehension training (ES 2.70). =
At least three qualifications are neces-
recognition measures, the magnitude She contrasted two training conditions,
sary, related to the above conclusions.
of change was .19 (on the Word Iden- one in which DI and main idea iden-
First, some may argue that our inclu-
tification subtest of the Woodcock tification were paired with self-
sion criteria for studies in the present
Reading Mastery Test) when com- monitoring training, and one in which synthesis were not restrictive enough-
pared to the control conditions. only DI and main idea identification
that is, only studies that reported stan-
There were, of course, some studies were provided. These instructional
dardized IQ and reading scores should
that were methodologically sophisti- conditions, in turn, were contrasted with have been included in the analysis.
cated that were not included in our a control, or no training, condition.
We do argue, however, that if we had
synthesis. Vellutino et al. (1996) were Graves found that the training con-
been unduly restrictive in how stud-
successful in remediating at-risk read- dition of combined DI plus self-
ies defined LD, we would have elimi-
ers over 15 weeks of 30-minute, daily, monitoring produced the best results nated approximately 70% of the stud-
one-to-one tutoring emphasizing pho- in identifying main ideas. This find-
ies in our data pool. For example, if
nological awareness, the alphabetic ing was subsequently replicated by we had considered only published,
principle, sight word vocabulary, and Graves and Levin (1989). Wong and
comprehension strategies. However, Jones also attempted to enhance the group-design studies reporting stan-
dardized IQ and reading scores in their
when we analyzed that particular reading comprehension of students
study we were unable to calculate ef- with LD through a metacognitive pro- sample, only 17 studies in the reading
fect sizes (see Table 2 of the original gram (ES .56). Six eighth and ninth
= recognition domain and 18 in the read-
article). That is, the one-to-one, inten- graders with LD were compared with ing comprehension domain would
sive tutoring, when compared to two a reading-matched control group of
have been included in the synthesis.
other conditions (traditional remedial sixth graders. Half of the students in Unfortunately, this narrowing of the
studies would not have provided an
programs in the schools), did not yield both the LD and the sixth-grade
effect sizes that were computable from accurate picture of the intervention
groups given self-questioning
were
the information presented. Indirectly, research that exists within special edu-
training procedures through which they
we could compute from the tables ef- learned to monitor their understand- cation for these particular students.
fect sizes on good and poor readers ing of text; the other half served as a Second, very few state departments
of education or research organizations
among the three conditions, two of control group. In the experimental con-
with technical reports.
which were variations of the school dition, the steps in comprehension provided us

approach of tutoring and classroom monitoring included probing ques- Therefore, we cannot generalize our

remediation, compared to the experi- tions, such as, &dquo;What are you study- findings beyond articles and disserta-
mental treatment of intensive one-to- ing this passage for?,&dquo; &dquo;Find the main tions. Although data were readily
one instruction. The effect sizes were idea in the paragraph and underline available across categories of children
less than .10. it,&dquo; &dquo;Think of a question about the placed in special education in unpub-
In sum, the results merely suggested main idea,&dquo; and so on. Their results lished reports ordepartment
state
that programs that emphasize seg- indicated that training increased the documents, department or
no state

menting information, such as in phon- awareness of important textual units research organization provided treat-
ics instruction, are not sufficient for bol- for adolescents with LD. ment data (i.e., means and standard
stering real-word recognition. There Surprisingly, we did not find that deviations) that focused primarily on
have been other reviews suggesting &dquo;control of task difficulty&dquo; contributed students with LD.
that phonological awareness is an significant variance to ES estimates. It Finally, we placed only a few re-

important, but not necessarily a suffi- could be that control of task difficulty, strictive parameters on how the pri-
 525

mary authors of the intervention stud- ologies flaws (e.g., no controls, inabil- for their assistance in data entry and cod-
ies defined LD. We initially relied on ity to calculate effect sizes). ing, and to Kathy Wilson for critical com-
whether studies &dquo;reported&dquo; that stu- ments on an earlier draft of this manuscript.

dents with LD or students with simi- The author is particularly appreciative of

lar labels (e.g., dyslexia, reading Summary extensive comments by three anonymous
reviewers of this journal and Drs. Ellen
disability, educational handicap) were Our synthesis characterized reading
Schiller, Harris Cooper, Bernice Wong, Phil
in the sample and were provided treat- intervention research within the field
Winne, Steve Forness, Sharon Vaughn, Ed
ment above and beyond their class- of LD over the last 30 years. We iden- Kameenui, and Reid Lyon on an earlier draft
room placement. Thus, our synthesis tified treatments that yielded high ef-
of this synthesis. The views of this report
was clearly biased toward studies that fect sizes for the domains of word are my own and do not necessarily reflect
use previously defined children or recognition and reading comprehen- those of the reviewers.
adults classified as learning disabled sion. A combined DI-SI model appears 4. Other publications related to this database
by the schools, clinics, and so forth. robust for reading comprehension, include Swanson and Hoskyn (1998) and
This criterion was refined in two ways. and a DI-only model appears robust Swanson and Sachse-Lee (in press).

First, we removed studies that did not for word recognition measures. In spite
state that the sample was in the aver- of the limitations in participant de-
NOTES
age IQ range. We also noted that sev- scriptions, we were also able to iden-
eral studies that did report IQ (ap- tify some aptitude variables (intel- 1. Although Cohen’s criteria were used to judge
proximately 35) included participants lectual range and severity of reading the magnitude of potential outcomes, the .20
with intelligence scores below 85 and deficiency) that played an important criterion may appear arbitrary to the reader.
therefore those samples were more ap- role in predicting treatment outcomes. Lipsy and Wilson (1993), in their review of
propriately classified as mildly retarded Studies that met cutoff score criteria the efficacy of psychological, educational,
or slow learners. On the basis of this bolstered the influence of direct in- and medical research, indicated that the
preliminary analysis, we implemented struction on word recognition out- grand mean for treatments related to 156
the criterion that the studies must state comes. Studies that included group meta-analysis and more than 1 million par-
that the sample is in the average range samples with discrepancies in IQ and ticipants was .47 for control-comparison de-
of intelligence. Therefore, a general reading but whose IQs were in the signs. Previous syntheses (see Lipsy & Wil-
son, 1993, for a review) placed reading
stipulation in the studies we synthe- high-average range, yielded lower instruction for elementary students in the
sized was that the sample-description effect sizes across treatments than
.60 range and program instruction influ-
section of each study stated that stu- samples with low-average IQs. ence on achievement in the -.10 to .29 range
dents with LD were in the normal (see Lipsy & Wilson’s Table 1, pp. 1186,
range of intelligence. 1189). Our recent meta-analysis of all do-
ABOUT THE AUTHOR
Our second refinement entailed mains (Swanson & Hoskyn, in press) placed
implementing a computer search of H. Lee Swanson, PhD, is a professor of edu-
overall educational treatment effectiveness
articles that emerged for the combined cational psychology and special education at in the .60 to .80 range. Although not a
terms high risk and treatment, or the the University of California. His research in- synthesis of interventions specifically, Satz
terms intervention and instruction. We et al. (1997) found, in their comprehensive
terests are working memory, intellectual de-
review of children and adolescents with mild
conducted a hand search of the top velopment, and problem solving as it applies head injury, that when compared to control
five journals over a 30-year period to to students with learning disabilities. Address:
H. Lee Swanson, School of Education, Univer- groups, outcomes for the former across time
locate intervention studies that in- on achievement measures supported the null
cluded samples of children considered sity of California-Riverside, CA 92521.
hypothesis. Thus, we reasoned (perhaps in-
at risk for LD who meet two psycho-
correctly) that if some academic domains
metric standards. These studies of AUTHOR’S NOTES require intensive intervention because of
&dquo;nonidentified&dquo; students with LD were this population’s unique processing prob-
included in our synthesis only if mean 1. This study was supported by U.S. Depart- lems (and as reflected in the effect sizes by
IQs of the sample were reported to be ment of Education Grant No. H023E40014, recent intervention studies, e.g., Torgesen
above 84 (or if it was reported that the Chesapeake Institute, and Peloy Endow- et al., 1997; Vellutino et al., 1996), then
in the average range ment Funds awarded to the author. The findings around .20 would be anticipated
participants were
views of this report do not necessarily re-
based on standardized scores) and a for short intervention periods. For medical
flect those of the Department of Education treatments related to life-threatening issues,
mean standardized reading score of
or the Chesapeake Institute. effective treatments are in the .08 to .20
the sample was reported to be at or
2. The work is a subsection of a larger synthe- range (Lipsy & Wilson, 1993), and there-
below the 25th percentile. Thirty-four sis submitted as a final report to the U.S. .20 is trivial finding.
fore not a
studies were located; 5 were included 2. At the present time, the controversy re-
Department of Education.
in the synthesis, and the remainder 3. I am most indebted to two research associ- garding the best reading instruction for stu-
were eliminated because of method- ates, Carole Sachse-Lee and Maureen Hoskyn, dents with LD is sharply polarized and
526

requires a less biased appraisal that respects We also analyzed studies not included in as a high-saturation vs. low-saturation
the risks associated with Type I and Type II our synthesis that included high-risk sam- condition) by giving credit if only one state-
errors. Thus, studies that included word ples. An intervention study that most di- ment fit within an instructional component
recognition and comprehension measures rectly addressed the issue of "explicitness" category rather than several statements. In
were the focus in this synthesis. These in- was Foorman et al.’s (1998). In that study, addition, we weighted each study via a meth-
cluded studies that directly focused on read- comparisons were made between first and odological composite score that took into

ing, as well as studies where instruction second graders receiving Title 1 services consideration internal validity, the amount
was not specific to reading. In this way, we for one of three kinds of classroom reading of instruction that varied from the control
could identify common instructional com- programs-direct instruction in letter- group, treatment integrity, and so on.
ponents that influenced reading regardless sound correspondences, practice in decodable 3. We assume a priori that students with LD
of the orientation of the study. text (explicit training in direct coding), have normal intelligence but suffer specific
We recognize, however, that conventional and less direct instruction in systematic achievement difficulties not directly related
wisdom suggests that readers with LD bene- sound-spelling patterns embedded in con- to environmental variables, such as SES and
fit most from reading programs that di- nected text (embedded code)-and two con- faulty teaching; thus, poor levels of achieve-
rectly emphasize reading skills (i.e., word trol conditions that involved the implicit ment are unexpected but may be explained
recognition), rather than more general read- instruction in the alphabetic coded while by neurological and/or cognitive correlates.
ing (or nonspecific reading) strategies. One reading connected text. Without correcting The reader may question, however, why we
reviewer of this manuscript also suggested for sample size and methodological varia- selected an overall mean IQ score of 85 and
that studies vary in their explicitness of in- tions, the effect size at the end of Grade 2 above as a criterion for article inclusion.
struction (e.g, direct instruction treatment for real-word recognition between the two That is, we may have been too restrictive in
explicitly focused on improving word recog- treatments that directly manipulated ex- our criterion. Besides the fact that 85 reflects

nition, or direct instruction focused on com- plicitness (direct vs. embedded, calculated 1 standard deviation below the mean, two
prehension), and this must be taken into from Table 4, p. 44) was .10. Both treat- additional reasons underlie this criterion.
consideration. ments were better than control conditions First, in the standardization of the most
In response to these concerns, we feel (.21 and .46). Thus, the magnitude of the popular intelligence measure for identify-
that separating studies by how well the treat- effect sizes (.10) did not warrant analyzing ing LD samples (Wechsler series), mean
ment was related to the dependent measure only experimental studies that included Full Scale IQs included in the larger stan-
is problematic for several reasons. First, it explicit training. What seems important, dardization sample for samples with LD
does not seem to be warranted from the ex- however, is that treatment included the seg- and those with RD were 93.8 and 99.8, re-
perimental literature. For example, in our mentation of words, as well as how system- spectively (see the WISC-III [1991] manual,
preliminary analysis we analyzed studies atic (based on the type of instructional Table 6.21, p. 213). An earlier comprehen-
that received high methodological compos- components) and methodologically sophis- sive (94 studies) meta-analysis of intelli-
ite scores (studies with high internal valid- ticated (internally valid) the intervention gence and learning disabilities by Kavale
ity). An exceptionally well-detailed study was. and Forness (1994) placed the mean com-
by Foorman et al. (1997) that overcame many Second, and more important, meta- posite IQ for students with LD at 97.01
of the artifacts in other studies suggested analysis is a procedure that allows for "cross- (see their Table 2, p. 11). Thus, we were
that the degree of explicit instruction re- study" inferences. To achieve stochastically liberal in our cutoff criteria. Second, low
lated to word segmentation may not produce dependent effect sizes, meta-analysis requires IQ confounds the distinctions among slow
clear differences on measures related to real combining effect sizes across multiple ex- learners, students with mild retardation,
word recognition (also see Lovett et al., perimental treatments within a study. Thus, and students with LD. As stated by Kavale
1990). Foorman et al. compared three in- because explicitness of instruction is ma- and Forness, in the most comprehensive
terventions (synthetic phonics [Orton- nipulated across multiple treatment within analysis of intelligence and LD to date,
Gillingham approach], analytic phonics a study, subtle effects of instruction may be

[focus on onsets and rimes], and sight word lost. Although one can safely assume that Intelligence has long been associated with
programs) using second and third graders these multiple treatments vary in terms of the MR concept, and it appears that any
identified with reading disabilities for explicitness, for statistical reasons they are suggestion to lower the IQ standard for
60 minutes a day for 1 school year. For viewed as instances ofa common treatment LD probably means greater confounding
word reading (an experimental measure of construct (see Gleser & Olkin, 1994, for a between LD and MR concepts. The pri-
words selected from first- and second-grade discussion). mary difficulty appears to surround stu-
cumulative vocabulary lists in which half Finally, inherent biases exist within some dents with IQs in the 70-85 range who
the words included predictable patterns and treatments because of their "overabundance" represent the "slow learner" group. This
half unpredictable or exceptional spelling of detail as compared to other treatments 13% of the population has long posed a
patterns), no significant differences emerged within a study. This was especially prob- dilemma. In all likelihood, their primary
between treatment conditions (see their Ta- lematic when a study failed to provide mea- manifestation will be low achievement,
ble 8, p. 270). The estimated effect sizes sures of treatment integrity (the degree to which given their IQ levels, is not unex-
varied from 0 to .18 for the various com- which each component of instruction is faith- pected. (p. 52)
parisons. Thus, the small magnitude in dif- fully administered as intended). We at-
ferences between treatments did not warrant, tempted to control for the fact that some We also recognize that there has been
in our opinion, a focus on the explicit de- studies provided more information about debate about the relevance of IQ, and of
tails of what was taught. treatment than others (we referred to this discrepancies between intelligence and read-
 527

ing, to the diagnosis of LD (see Stanovich subtle differences among literacy variables policies and scientific research. Reading
& Siegel, 1994, for a comprehensive review). in the experimental literature were treat- and Writing: An Interdisciplinary Journal,
To determine if differences between intelli- ment comparisons within studies, whereas 5, 113-139.
gence and reading influence outcomes, we meta-analyses focused on comparisons across Becker, W., & Carnine, D. (1980). Direct
allowed for the possibility that samples might studies. In addition, the categories of lit- instruction: An effective approach for
have mean IQ and reading scores in the eracy measures (dependent measures) within educational intervention with the dis-
same low range (85 to 90), as well as large studies (e.g., inferential and literal compre- advantaged and low performers. In B.
mean score differences between IQ and read- hension) were not statistically independent. Lahey & A. Kazdin (Eds.), Advances in
ing. This allowed for a testing of whether This lack of statistical independence within child clinical psychology. New York: Ple-
various combinations of IQ and reading studies required aggregating dependent mea- num Press.
moderated treatment outcomes. sures (Hunter & Schmidt, 1994). Berninger, V. W. (1994). Reading and writ-
4. As the reader can easily tell, there is over- Second, because of the bias related to posi- ing acquisition: A developmental neuro-
lap in the listing of activities that reflect tive outcomes in the synthesis of treatments psychological perspective. Madison, WI:
certain instructional components (more than (e.g., Lipsey & Wilson, 1993; see also Weisz Brown & Benchmark.
1 of the 45 codes can occur in several cat- et al.’s [1995] discussion of the "dodo ver- Bishop, D. V. M., & Adams, C. (1990). A
egories; e.g., see Nos. 3 and 11). Treatments dict"), it was critical that the commonali- prospective study of the relationship
were coded as reflecting 1 of 19 compo- ties of effective treatment were identified. between specific language impairment,
nents (Component 5 required two activi- That is, we could not adequately assess the phonological disorder and reading re-
ties) if at least one of the appropriate "what" of instruction unless we clearly iden- tardation. Journal of Child Psychology and
instructional activities was stated in the tified the "how." As shown in this synthe- Psychiatry, 31, 1027-1050.
treatment section. However, once an in- sis, there are tremendous differences in Bollen, K. A. (1989). Structural equations
structional activity was credited to a com- instructional activities, as well as a host of with latent variables. New York: Wiley-
ponent category, that same instructional other methodological variables, that moder- Interscience.
activity could not be used for another in- ate treatment outcomes. Unless these vari- Borkowski, J. G., Estrada, M. T., Milstead,
structional component. In rare cases where ables are identified and their influence on out- M., & Hale, C. A. (1989). General problem-
comes clearly delineated, testing the subtle
only one instructional activity was stated solving skills: Relations between meta-
and that instructional activity could be ap- aspects of literacy becomes a moot point. cognition and strategic processing. Learn-
plied to more than one category, preference 5. Because several studies may have been in- ing Disability Quarterly, 12, 57-70.
was given to the categories 1, 2, 3, 8, and advertently left out of our synthesis (in addi- Borkowski, J. G., & Turner, L. A. (1990). Trans-
17. These categories reflect a basic instruc- tion to the fact that published studies are situational characteristics of metacogni-
tional core outlined by Rosenshine and Ste- often biased), we calculated the number of tion. In W. Schneider & F. Weinert (Eds.),
vens (1986) and Slavin, Stevens, and Mad- studies necessary (i.e., studies that reported Interactions among aptitudes, strategies, and
den (1988) shared by the DI and SI models. null results) that would threaten our over-
knowledge (pp. 159-176). New York:
The categories were not intended to be mu- all conclusions (i.e., the File Drawer Prob-
Springer Verlag.
tually exclusive, however, but rather to cap- lem). We calculated the tolerance levels for Borkowski, J. G., Weyhing, R. S., & Carr,
ture an array of components identified in null results (see Hunter & Schmidt, 1990, M. (1988). Effects of attributional retrain-
the literature. Table 22.2, for the number of null results
ing onstrategy-based reading compre-
as a function of the number of studies). The hension in
Although our codes were based on reviews learning disabled students.
number of studies required to indicate an
of instructional literature, one reviewer was Journal of Educational Psychology, 80, 46-
critical because we did not code the inde- inclusion bias for this synthesis was 191 for 53.
pendent variables by what aspect of liter- reading comprehension studies and 75 for Brophy, J., & Good, T. (1986). Teacher-
word recognition studies. Thus, 191 read-
acy was addressed (e.g., phonological effects results. In M. C. Wittrock (Ed.),
awareness, spelling-sound relationships, ing comprehension studies and 75 word rec- Handbook of research on teaching (3rd ed.).
word identification, sentence structure, lit- ognition studies meeting the synthesis cri- New York: Macmillan.
eral conprehension, inferential comprehen- teria and reporting null results would have
Bruck, M. (1990). Word-recognition skills
to have been not retrieved before one could
sion, etc.) but instead focused on how it of adults with childhood diagnoses of
was taught. Although the meager number
conclude that our selection of studies re-
dyslexia. Developmental Psychology, 26,
of experimental studies that met our crite- flected a sampling bias. 439-454.
rion for internal validity limited some of Byrne, B., Freebody, P., & Gates, A. (1992).
our analysis, we focused on how the inde- Longitudinal data on relations of word-
REFERENCES
pendent variable was delivered for two rea- reading strategies to comprehension,
sons. First, although the literacy data were Aaron, P. G., Frantz, S. S., & Manges, A. R. reading time, and phonemic awareness.
available to us, for statistical reasons the (1990). Dissociation between comprehen- Reading Research Quarterly, 27, 140-151.
experimental conditions were collapsed in sion and pronunciation in dyslexic and Cohen, J. (1988). Statistical power analysis
the analysis (see Hunter & Schmidt, 1994). hyperlexic children. Reading & Writing, for the behavioral sciences (2nd ed.). New
As previously stated, multiple-treatment , 243-264.
2 York: Academic Press.
studies (e.g., those that systematically var- Adams, M. J. (1990). Beginning to read. Cam- Cooper, H., Nye, B., Charlton, K., Lindsey,
ied the degree of word segmentation during bridge, MA: MIT Press. J., & Greathouse, S. (1996). The effects of
instruction) were regarded as instances of Adams, M. J., & Bruck, M. (1993). Word summer vacation on achievement scores:
a common treatment construct. Most of the recognition: The interface of educational A narrative and meta-analysis review.
528

Review of Educational Research, 66, 227- Graham, S., & Harris, K. R. (1993). Cogni- prehension or word recognition disabili-
268. tive strategy instruction: Methodologi- ties. Journal of Reading Behavior, 11
(1), 41-
Deshler, D. D., & Schumaker, J. B. (1988). cal issues and guidelines in conducting 48.
An instructional model for teaching stu- research. In S. Vaughn & C. Bos (Eds.), Lipsey, M. W., & Wilson, D. B. (1993). The
dents how to learn. In J. Graden, J. Zins, Research issues in learning disabilities efficacy of psychological, educational,
& M. Curtis (Eds.), Alternative educational (pp. 146-158). New York: Springer-Verlag. and behavioral treatment: Confirmation
delivery systems: Enhancing instructional Graves, A. W., & Levin, J. R. (1989). Com- from meta-analysis. American Psycholo-
options for all students (pp. 391-411). Wash- parison of monitoring and mnemonic gist, 48, 1181-1209.
ington, DC: National Association of text-processing strategies in learning dis- Marks, M., Pressley, M., Coley, J. D., Craig,
SchoolPsychologists. abled students. Learning Disability Quar- S., Gardner, R., Rose, W., & DePinto, T.
Engelmann, S., & Carnine, D. W. (1982). terly, 12, 212-236. (1993). Teachers’ adaptations of recipro-
Theory of instruction: Principles and appli- Guthrie, J. T., VanMeter, P., McCann, A. D., cal teaching: Progress toward a classroom-
cations. New York: Irvington. & Wigfield, A. (1996). Growth of literacy compatible version of reciprocal teach-
Ehri, L. C. (1994). Development of the abil- engagement: Changes in motivations and ing. Elementary School Journal, 94, 267-283.
ity to read words: Update. In R. Ruddell, strategies. Reading Research Quarterly, 31, Meichenbaum, D. (1977). Cognitive behav-
M. Ruddell, & H. Singer (Eds.), Theoreti- 306-332. ior modification. New York: Plenum Press.
cal models and processes of reading (pp. 323- Harris, K. R., & Pressley, M. (1991). The Morrison, S. R., & Siegel, L. S. (1991). Learn-
358). Newark, DE: International Read- nature of cognitive strategy instruction: ing disabilities: A critical review of defi-
ing Association. Interactive strategy construction. Excep- nitional and assessment issues. In J. E.
Fletcher, J., Shaywitz, S. E., Shankweiler, tional Children, 57, 392-404. Obrzut & G. W. Hynd (Eds.), Neuro-
D. P., Katz, L., Liberman, I., Stuebing, Harwell, M. R. (1992). Summarizing Monte logical foundations of learning disabilities
K., Francis, D., Fowler, A., & Shaywitz, Carlo results in methodological research. (pp. 79-97). San Diego: Academic Press.
B. A. (1994). Cognitive profiles of read- Journal of Educational Statistics, 17, 297- Olson, R. K., Wise, B., Johnson, M., & Ring,
ing disability: Comparisons of discrepancy 313. J. (1997). The etiology and remediation
and low achievement definitions. Jour- Hedges, L. V. (1994a). Statistical consider- of phonological-based word recognition
nal of Educational Psychology, 86, 6-23. ations. In H. Cooper & L. V. Hedges and spelling disabilities: Are phonologi-
Foorman, B. R. (1994). The relevance of a (Eds.), The handbook of research synthesis cal deficits the "hole" story? In B. A. Blach-
connectionist model of reading for "the (pp. 29-38). New York: Russell Sage. man (Ed.), Foundation of reading acquisi-

great debate." Educational Psychology Hedges, L. V. (1994b). Fixed effects mod- tion (pp. 305-326). Mahwah, NJ: Erlbaum.
Review, 6(1), 25-45. els. In H. Cooper & L. V. Hedges (Eds.), Palincsar, A. S. (1986). The role of dialogue
Foorman, R., Francis, D. J., Fletcher, J. M.,
B. The handbook of research synthesis (pp. 285- in providing scaffolded instruction. Edu-
& Lynn, A. (1996). Relation of phono- 299). New York: Russell Sage. cational Psychologist, 21
(1 & 2), 73-98.
logical and orthographic processing to Hedges, L. V., & Olkin, I. (1985). Statistical Palincsar, A. S., & Brown, A. L. (1984).
early reading: Comparing two ap- methods for meta-analysis. San Diego, CA: Reciprocal teaching of comprehension-
proaches to regression-based, reading- Academic Press. fostering and comprehension monitor-
level-match design. Journal of Educational Hunter, J. E., & Schmidt, F. L. (1990). Methods ing activities. Cognition and Instruction,
Psychology, 8, 639-652. of meta-analysis: Correcting error and bias 1, 117-175.
Foorman, B. R., Francis, D. J., Fletcher, J. M., in research findings. Newbury Park, CA: Palincsar, A. S., & Brown, A. L. (1988).
Schatschneider, C., & Mehta, P. (1998). Sage. Teaching and practicing thinking skills
The role of instruction in learning to read: Hunter, J. E., & Schmidt, F. L. (1994). Cor- to promote comprehension in the con-

Preventing reading failure in at-risk chil- recting for sources of artificial variation text of group problem solving.Remedial
dren. Journal of Educational Psychology, across studies. In H. Cooper & L. V. and Special Education, 9(1), 53-59.
(1),
90 37-55. Hedges (Eds.), The handbook of research Pflaum, S. W. (1980). The predictability of
Francis, D. J., Shaywitz, S. E., Stuebing, synthesis (pp. 323-336). New York: Rus- oral reading behaviors on comprehen-
K. K., Shaywitz, B. A., & Fletcher, J. M. sell Sage. sion in learning disabled and normal
(1996). Developmental lag versus deficit Kameenui, E. J., Jitendra, A. K., & Darch, readers. Journal of Reading Behavior, 12,
models of reading disability: A longitu- C. B. (1995). Direct instruction reading 231-236.
dinal, individual growth curves analy- as contronym and eonomine. Reading & Pinnell, G. S., Lyons, C. A., DeFord, D. E.,
sis. Journal of Educational Psychology, 88, Writing Quarterly: Overcoming Learning Bryk, A. S., & Seltzer, M. (1994). Com-
3-17. (1), 3-17.
Difficulties, 11 paring instructional models for the lit-
Frederiksen, J. R., & Warren, B. M. (1987). Kavale, K. A., & Forness, S. R. (1994). Learn- eracy education of high-risk first graders.
A cognitive framework for developing ing disabilities andintelligence. In T. Reading Research Quarterly, 29, 8-39.
expertise in reading. In R. Glaser (Ed.), Scruggs & Mastropieri (Eds.), Ad-
M. Pressley, M., Brown, R., El-Dinary, P. B.,
Advances in instructional psychology vances in learning disabilities and behav- & Allferbach, P. (1995). The comprehen-
(Vol. 3, pp. 1-39) Hillsdale, NJ: Erlbaum. ioral disabilities (Vol. 8, pp. 1-63). Green- sion instruction that students need: In-
Gleser, L. J., & Olkin, I. (1994). Stochasti- wich, CT: JAI Press. struction fostering constructively respon-
cally dependent effect sizes. In H. Coo- Kendall, J. R., & Hood, J. (1979). Investi- sive reading. Learning Disabilities Research &
per & L. V. Hedges (Eds.), The handbook gating the relationship between compre- Practice, 10, 215-224.
of research synthesis (pp. 339-355). New hension and word recognition: Oral Pressley, M., El-Dinary, P. B., Gaskins, I.,
York: Russell Sage. reading analysis of children with com- Schuder, T., Berrgman, J. L., Almasi, J.,
 529

& Brown, R. (1992). Beyond direct ex- Siegel, L. S. (1993). The cognitive basis of learning disabilities. School Psychology Re-
planation: Transactional instruction of dyslexia. In M. Howe & R. Pasnak (Eds.), view, 25, 370-391.
reading comprehension strategies. Ele- Emerging themes in cognitive development Swanson, H. L., & Hoskyn, M. (1998). Ex-
mentary School Journal, 92, 513-556. (pp. 33-52). New York: Springer Verlag. perimental intervention research for stu-
Pressley, M., & Ghatala, E. S. (1990). Self- Siegel, L. S., & Ryan, E. B. (1989). The de- dents with learning disabilities: A
regulated learning: Monitoring learning velopment of working memory in nor- meta-analysis of treatment outcomes.
from text. Educational Psychologist, 25, 19-34. mally achieving and subtypes of learning Review of Educational Research, 68, 277-321.
Pressley, M., & Harris, K. R. (1994). In- disabled children. Child Development, 60, Swanson, H. L., with Hoskyn, M., & Lee, C.
creasing the quality of educational inter- 973-980. (1999). Interventions for students with learn-
vention research. Educational Psychology Slavin, R. E., Karweit, N. L., & Madden, ing disabilities: A meta-analysis of treatment
Review, 6, 191-208. N. A. (1989). Effective programs for stu- outcomes. New York: Guilford.
Pressley, M., & Rankin, J. (1994). More dents at risk. Needham Heights, MA: Swanson, H. L., & Sachs-Lee, C. (in press).
about whole language methods of read-
Allyn & Bacon. A meta-analysis of single-subject design
ing instruction for students at risk for intervention research for students with
Slavin, R. E., Stevens, R. J., & Madden, N. A.
early reading failure. Learning Disabili- (1988). Accommodating student diversity LD. Journal of Learning Disabilities.
ties Research & Practice, 9, 157-168. in reading and writing instruction: A co- Vellutino, F., & Scanlon, D. M. (1991). The
Resnick, L. B. (1987). Constructing knowl-
operative learning approach. Remedial and effects of instructional bias on word iden-
edge in school. In L. S. Liben (Ed.), De- Special Education, 9(1), 60-66. tification. In I. L. Rieben & C. A. Perfetti
velopment and learning: Conflict or congru- Spector, J. (1995). Phonemic awareness
E. (Eds.), Learning to read: Basic research and
ence? (pp. 19-50). Hillsdale, NJ: Erlbaum.
training: Application of principles of its implication (pp. 189-204). Hillsdale,
Reynolds, C. R. (1981). The fallacy of "two direct instruction. Reading & Writing NJ: Erlbaum.
years below grade level for age" as a
Quarterly: Overcoming Learning Difficul- Vellutino, F., Scanlon, D. M., Sipay, E.,
diagnostic criterion for reading disorders. (1), 37-51.
ties, 11 Small, S., Pratt, A., Chen, R., & Denckla,
Journal of School Psychology, 11, 250-258.
Rosenshine, B. (1995). Advances in research Stanovich, K. E. (1986). Matthew effects in M. (1996). Cognitive profiles of difficult-
on instruction. Journal of Educational Re- reading: Some consequences of indi- to-remediate and readily remediated
vidual differences in the acquisition of poor readers: Early intervention as a
search, 88, 262-268.
Rosenshine, B., & Stevens, R. (1986). Teach- literacy. Reading Research Quarterly, 21, vehicle for distinguishing between cog-
360-406. nitive and experiential deficits as basic
ing functions. In M. C. Wittrock (Ed.), causes of specific reading disability. Jour-
Handbook of research on teaching (3rd ed.). Stanovich, K. E., & Siegel, L. S. (1994).
Phenotypic performance profile of nal of Educational Psychology, 88, 601-638.
New York: Macmillan.
children with reading disabilities: A Wechsler, D. (1991). Wechsler intelligence
Satz, P., Zaucha, K., McCleary, C., Light,
R., Asarnow, R., & Becker, B. (1997). Mild regression-based test of the phonological- scale for children-III (administration
core difference model. Journal of Educa- manual). San Antonio, TX: Psychologi-
head injury in children and adolescents:
A review of studies tional Psychology, 86, 24-53. cal Corp.
(1970-1995). Psycho-
logical Bulletin, 122,107-131. Swanson, H. L. (1991). Operational defini- Weisz, J. R., Weiss, B., Han, S. S., Granger,
Shankweiler, D., & Crain, S. (1986). Language tions of learning disabilities. Learning D. A., & Morton, T. (1995). Effects of
mechanisms and reading disorder: A mod- Disability Quarterly, 14, 242-254. psychotherapy with children and ado-
ular approach. Cognition, 24, 139-168. Swanson, H. L., & Alexander, J. E. (1997). lescents revisited: A meta-analysis of
Share, D. L., & Stanovich, K. E. (1995). Cognitive processes as predictors of word treatment outcome studies. Psychologi-

Cognitive processes in early reading recognition and reading comprehension cal Bulletin, 117, 450-468.
development: Accommodating indi- in learning-disabled and skilled readers: Wong, B. Y. L. (1991). Assessment of meta-
vidual differences into a model of ac- Revisiting the specificity hypothesis. Jour- cognitive research in learning disabili-
quisition. Issues In Education, 1, 1-58. nal of Educational Psychology, 89, 128-158. ties: Theory, research, and practice. In
Siegel, L. S. (1992). An evaluation of the Swanson, H. L., Carson, C., & Sachse-Lee, H. L. Swanson (Ed.), Handbook on the as-
discrepancy definition of dyslexia. Jour- C. M. (1996). A selective synthesis of in- sessment of learning disabilities (pp. 265-
nal of Learning Disabilities, 25, 618-629. tervention research for students with 284). Austin, TX: PRO-ED.
530

APPENDIX
References for theMeta-Analysis

*Reading comprehension studies. a specific neurodevelopmental therapy for the remediation

**Word recognition studies. of learning disabilities. Developmental and Behavioral Pedi-
***Combined comprehension and recognition studies. atrics, 5, 189-194.
***14. Cartelli, L. M. (1978). Paradigmatic language train-
**1. Ayres, A. J. (1972). Improving academic scores through ing for leaning disabled children. Journal of Learning Disabil-
sensory integration. Journal of Learning Disabilities, 5, 338- ities, 11(5), 54-59.
343. *15. Chan, L. K. S. (1991). Promoting strategy generali-
**2. Bakker, D. J., Bouma, A., & Gardien, C. J. (1990). zation through self-instructional training in students with read-
Hemisphere-specific treatment of dyslexia subtypes: A field ing disabilities. Journal of Learning Disabilities, 24, 427-433.
experiment. Journal of Learning Disabilities, 23, 433-438. *16. Chan, L. K. S., & Cole, P. G. (1986). The effects of
**3. Balcerzak, J. P. (1986). The effects of an aptitude comprehension monitoring training on the reading compe-
treatment interaction approach with intermediate aged learn- tence of learning disabled and regular class students. Reme-
ing disabled students based on emphasizing the individual’s dial and Special Education, 7(4), 33-40.
strength in simultaneous or sequential processing in the *17. Chan, L. K. S., Cole, P. G., & Morris, J. N. (1990).
areas of mathematics, reading and self-concept. Disserta- Effects of instruction in the use of a visual-imagery strategy
tion Abstracts International, 47(3-A), 630. on the reading-comprehension competence of disabled and
*4. Bos, C. S., & Anders, P. L. (1990). Effects of interac- average readers. Learning Disability Quarterly, 13, 2-11.
tive vocabulary instruction on the vocabulary learning and *18. Chase, C. H., Schmitt, R. L., Russell, G., & Tallal, P.
reading comprehension of junior-high learning disabled stu- (1984). A new chemotherapeutic investigation: Piracetam
dents. Learning Disability Quarterly, 13, 31-42. effects on dyslexia. Annals of Dyslexia, 34, 29-48.
*5. Bos, C. S., & Anders, P. L. (1992). Using interactive **19. Cornelius, P. L., & Semmel, M. 1. (1982). Effects of
teaching and learning strategies to promote text comprehen- summer instruction on reading achievement regression of
sion and content learning for students with learning disabili- learning disabled students. Journal of Learning Disabilities,
ties. International Journal of Disability, Development and 15, 409-413.
Education, 39, 225-238. *20. Darch, C., & Carnine, D. (1986). Teaching content
*6. Bos, C. S., Anders, P. L., Filip, D., & Jaffe, L. E. (1985). area material to learning disabled students. Exceptional Chil-
Semantic feature analysis and long-term learning. National dren, 53, 240-246.
Reading Conference Yearbook, 34, 42-47. *21. Darch, C., & Eaves, R. C. (1986). Visual displays to
*7. Bos, C. S., Anders, P. L., Filip, D., & Jaffe, L. E. increase comprehension of high school learning-disabled
(1989). The effects of an interactive instructional strategy for students. The Journal of Special Education, 20, 309-318.
enhancing reading comprehension and content area learn- *22. Darch, C., & Gersten, R. (1986). Direction-setting
ing for students with learning disabilities. Journal of Learning activities in reading comprehension: A comparison of two
Disabilities, 22, 384-390. approaches. Learning Disability Quarterly, 9, 235-243.
**8. Branwhite, A. B. (1983). Boosting reading skills by *23. Darch, C., & Kameenui, E. J. (1987). Teaching LD
direct instruction. British Journal of Educational Psychology, students critical reading skills: A systematic replication. Learn-
53, 291-298. ing Disability Quarterly, 10, 82-90.
**9. Brown, I. S., & Felton, R. C. (1990). Effects of instruc- **24. Das, J. P., Mishra, R. K., & Pool, J. E. (1995). An
tion on beginning reading skills in children at risk for reading experiment on cognitive remediation of word-reading diffi-
disability. Reading & Writing: An Interdisciplinary Journal, 2, culty. Journal of Learning Disabilities, 28, 66-79.
223-241. ***25. DeBoskey, D. S. (1982). An investigation of the reme-
**10. Brown, R. T., & Alford, N. (1984). Ameliorating diation of learning disabilities based on brain-related tasks
attentional deficits and concomitant academic deficiencies in as measured by the Halstead-Reitan Neuropsychological
learning disabled children through cognitive training. Journal Test Battery. Dissertation Abstracts International, 43(6-B),
of Learning Disabilities, 17, 20-26. 2032.
**11. Bryant, N. D., Fayne, H. R., & Gettinger, M. (1982). *26. Dixon, M. E. (1984). Questioning strategy instruction
Applying the mastery learning model to sight word instruc- participation and reading comprehension of learning disabled
tion for disabled readers. Journal of Experimental Education, students. Dissertation Abstracts International, 44(11-A), 3349.
50, 116-121. **27. Englert, C. S., Hiebert, E. H., & Stewart, S. R. (1985).
**12. Bryant, S. T. (1979). Relative effectiveness of visual- Spelling unfamiliar words by an analogy strategy. The Jour-
auditory versus visual-auditory-kinesthetic-tactile procedures nal of Special Education, 19, 291-306.
for teaching sight words and letter sounds to young, dis- *28. Englert, C. S., & Mariage, T. V. (1991 ). Making students
abled readers. Dissertation Abstracts International, 40(5-A), partners in the comprehension process: Organizing the reading
2588-2589. &dquo;Posse.&dquo; Learning Disability Quarterly, 14, 123-138.
**13. Carte, E., Morrison, D., Sublett, J., Uemura, A., & ***29. Farmer, M. E., Klein, R., & Bryson, S. E. (1992).
Setrakian, W. (1984). Sensory integration therapy: A trial of Computer-assisted reading: Effects of whole word feedback
 531

on fluency and comprehension in readers with severe dis- (1978). The comparative effectiveness
*46. Lerner, C. H.
abilities. Remedial and Special Education, 13, 50-60. of language experience approach and a basal-type approach
a
**30. Fiedorowicz, C. A. M. (1986). Training of component to remedial reading instruction for severely disabled readers
reading skills. Annals of Dyslexia, 36, 318-334. in a senior high school. Dissertation Abstracts International,
***31. Fiedorowicz, C. A. M., & Trites, R. L. (1987). An eval- 39(2-A), 779-7800.
uation of the effectiveness of computer-assisted component **47. Lorenz, L., & Vockell, E. (1979). Using the neuro-
reading subskills training. Ontario, Canada: Queen’s Printer. logical impress method with learning disabled readers. Jour-
**32. Foorman, B. R., Francis, D. J., Winikates, D., Mehta, nal of Learning Disabilities, 12, 420-422.
P., Schatshneider, C., & Fletcher, J. M. (1997). Early inter- ***48. Lovett, M. W., Borden, S. L., DeLuca, T., Lacerenza,
ventions for children with reading disabilities. Scientific Stud- L., Benson, N. J., & Brackstone, D. (1994). Treating the core
ies of Reading, 3, 255-276. deficits of developmental dyslexia: Evidence of transfer of
*33. Gajria, M., & Salvia, J. (1992). The effects of summa- learning after phonologically- and strategy-based reading
rization instruction on text comprehension of students with training programs. Developmental Psychology, 30, 805-822.
learning disabilities. Exceptional Children, 58, 508-516. **49. Lovett, M. W., Ransby, M. J., & Barron, R. W., (1988).
***34. Gittelman, R., & Feingold, 1. (1983). Children with Treatment, subtype, and word type effects on dyslexic
reading disorders: I. Efficacy of reading remediation. Journal children’s response to remediation. Brain and Language,
of Child Psychology and Psychiatry, 24, 167-191. 34(2), 328-349.
*35. Graves, A. W. (1986). Effects of direct instruction ***50. Lovett, M. W., Ransby, M. J., Hardwick, N., Johns,
and metacomprehension training on finding main ideas. Learn- M. S., & Donaldson, S. A. (1989). Can dyslexia be treated?
ing Disabilities Research, 1, 90-100. Treatment-specific and generalized treatment effects in dys-
***36. Helper, M. M., Farber, E. D., & Feldgaier, S. (1982). lexic children’s response to remediation. Brain and Language,
Alternative thinking and classroom behavior of learning 37(1 ), 90-121.
impaired children. Psychological Reports, 50, 415-420. **51. Lovett, M. W., & Steinbach, K. A. (1997). The effec-
***37. Humphries, T. W., Wright, M., Snider, L., & tiveness of remedial programs for reading disabled children
McDougall, B. (1992). A comparison of the effectiveness of of different ages: Is there decreased benefit for older chil-
sensory integrative therapy and perceptual-motor training in dren ? Learning Disability Quarterly, 20, 189-210.
treating children with learning disabilities. Developmental and **52. Lovett, M. W., Warren-Chaplin, P. M., Ransby, M. J.,
Beha vioral Pediatrics, 13(l), 31-40. & Borden, S. L. (1990). Training the word recognition skills
*38. Johnson, L., Graham, S., & Harris, K. R. (1997). The of reading disabled children: Treatment and transfer effects.
effects of goal setting and self-instructions on learning a Journal of Educational Psychology, 82, 769-780.
reading comprehension strategy: A study with students with *53. Lovitt, T., Rudsit, J., Jenkins, J., Pious, C., & Benedetti,
learning disabilities. Journal of Learning Disabilities, 30, 80-91. D. (1986). Adapting science materials for regular and learn-
**39. Jones, K. M., Torgesen, J. K., & Sexton, M. A. (1987). ing disabled seventh graders. Remedial and Special Educa-
Using computer guided practice to increase decoding flu- tion, 7, 31-39.
ency in learning disabled children: A study using the Hint ***54. Lucangeli, D., Galderisi, D., & Cornoldi, C. (1995).
and Hunt I Program. Journal of Learning Disabilities, 20, Specific and general transfer effects following metamemory
122-128. training. Learning Disabilities Research & Practice, 10(1 ),
**40. Kendall, P. C., & Braswell, L. (1982). Cognitive- 11-21.
behavioral self-control therapy for children: A components ***55. Lundberg, I., & Olofsson, A. (1993). Can computer
analysis. Journal of Consulting and Clinical Psychology, 50, speech support reading comprehension? Computers in Human
672-689. Behavior, 9(2-3), 283-293.
**41. Kennedy, K. M., & Backman, J. (1993). Effective- *56. MacArthur, C. A., & Haynes, J. B. (1995). Student
ness of the Lindamood Auditory Discrimination In-Depth Pro- Assistant for Learning from Text (SALT): A hypermedia reading
gram with students with learning disabilities. Learning Dis- aid. Journal of Learning Disabilities, 28, 150-159.
abilities Research & Practice, 8, 253-259. *57. Manning, B. H. (1984). Problem-solving instruction
**42. Kershner, J. R., Cummings, R. L., Clarke, K. A., as an oral comprehension aid for reading disabled third
Hadfield, A. J., & Kershner, B. A. (1990). Two-year evalua- graders. Journal of Learning Disabilities, 17, 457-461.
tion of the Tomatis Listening Training Program with learning **58. Maron L. R. (1993). A comparison study of the effects
disabled children. Learning Disability Quarterly, 13, 43-53. of explicit versus implicit training of test-taking skills for
**43. Kim, Y. O. (1992). The effect of teaching a test learning-disabled fourth-grade students. Dissertation Abstracts
taking strategy to high school students with learning disabili- International, 53(9-B), 4613.
ties. Dissertation Abstracts International, 53(1-A), 121. ***59. Mathes, P. G., & Fuchs, L. S. (1993). Peer-mediated
*44. Klingner, J. K., & Vaughn, S. (1996). Reciprocal teach- reading instruction in special education resource rooms. Learn-
ing of reading comprehension strategies for students with ing Disabilities Research & Practice, 8, 233-243.
learning disabilities who use English as a second language. *60. Meadows, N. B. W. (1989). The effect of individual,
Elementary School Journal, 96, 275-293. teacher directed and cooperative learning instructional meth-
*45. Leong, C. K., Simmons, D. R., & Izattt-Gambell, M. A. ods on and the comprehension of expository text. Disserta-
(1990). The effect of systematic training in elaboration on tion Abstracts International, 50(2-A), 406.
word meaning and prose comprehension in poor readers. ***61. Meyer, L. A. (1982). The relative effects of word-
Annals of Dyslexia, 40, 192-215. analysis and word-supply correction procedures with poor
532

readers during word-attack training. Reading Research Quar- *77. Snider, V. E. (1989). Reading comprehension perfor-
terly, 17, 544-555. mance of adolescents with learning disabilities. Learning
***62. Morgan, A. V. (1991). A study of the effects of Disability Quarterly, 12(2), 87-96.
attribution retraining and cognitive self-instruction upon the *78. Stevens, R. J., & Slavin, R. E. (1995). The coopera-
academic and attentional skills, and cognitive-behavioral tive elementary school: Effect on student achievement and
trends of elementary-age children served in self-contained social relations. American Educational Research Journal,
learning disabilities programs. Dissertation Abstracts Inter- 32, 321-351.
national, 51 (8-B), 4035. ***79. Sullivan, J. (1972). The effects of Kephart’s
**63. Naylor, J. G., & Pumfrey, P. D. (1983). The alleviation perceptual-motor training on a reading clinic sample. Journal
of psycholinguistic deficits and some effects on the reading of Learning Disabilities, 5, 545-551.
attainments of poor readers: A sequel. Journal of Research ***80. Swanson, H. L., & Trahan, M. F. (1992). Learning
in Reading, 6, 129-153. disabled readers’ comprehension of computer mediated text:
The influence of working memory, metacognition and attri-
***64. Olofsson, A. (1992). Synthetic speech and com-
bution. Learning Disabilities Research & Practice, 7, 74-86.
puter aided reading for reading disabled children. Reading & ***81. Torgesen, J. K., Wagner, R. K., Rashotte, C. A.,
Writing, 4, 165-178.
**65. Olson, R. K., & Wise, B. W. (1992). Reading on the
Alexander, A. W., & Conway, T. (1997). Preventive and
remedial interventions for children with severe reading dis-
computer with orthographic and speech feedback. Reading abilities. Learning Disabilities, 8(l), 51-61.
& Writing: An Interdisciplinary Journal, 4, 107-144.
**82. Van Daal, V. H. P., & Reitsma, P. (1990). Effects of
*66. Pany, D., Jenkins, J. R., & Schreck, J. (1982). Vocab-
independent word practice with segmented and whole-word
ulary instruction: Effects on word knowledge and reading sound feedback in disabled readers. Journal of Research in
comprehension. Learning Disability Quarterly, 5, 202-215. Reading, 13(2), 133-148.
**67. Porinchak, P. M. (1984). Computer-assisted instruc- **83. Van Daal, V. H. P., & Reitsma, P. (1993). The use of
tion in secondary school reading: Interaction of cognitive speech feedback by normal and disabled readers in com-
and affective factors. Dissertation Abstracts International, puter-based reading practice. Reading & Writing: An Inter-
45(2-A), 478. disciplinary Journal, 5, 243-259.
**68. Prior, M., Frye, S., & Fletcher, C. (1987). Remediation **84. Van Daal, V. H. P., & Van Der Leij, D. (1992).
for subgroups of retarded readers using a modified oral Computer-based reading and spelling practice for children
spelling procedure. Developmental Medicine and Child Neurol- with learning disabilities. Journal of Learning Disabilities, 25,
ogy, 29(1 ), 64-71. 186-195.
***69. Ratekin, N. (1979). Reading achievement of disabled *85. Wade, J. F. (1979). The effects of component deficit
learners. Exceptional Children, 45, 454-458. remediation and academic deficit remediation on improving
*70. Reilly, J. P. (1991 ). Effects of a cognitive-behavioral reading achievement of learning disabled children. Disserta-
program designed to increase the reading comprehension tion Abstracts International, 40(3-A), 1412.
skills of learning-disabled students. Dissertation Abstracts *86. Wade, J., & Kass, C. E. (1987). Component deficit
International, 52(3-A), 865. and academic remediation of learning disabilities. Journal of
*71. Schulte, A. C., Osborne, S. S., & McKinney, J. D. Learning Disabilities, 20, 441-447.
**87. Warner, J. M. R. (1973). The effects of two treatment
(1991 ). Academic outcomes for students with learning dis-
abilities in consultation and resource programs. Exceptional modes upon children diagnosed as having learning disabili-
ties. Dissertation Abstracts International, 34(3-A), 1142-1143.
Children, 57, 162-172.
**88. Waterman, D. E. (1974). Remediation of word attack
***72. Scruggs, T. E., Mastropieri, M. A., & Tolfa-Veit, D.
skills in slow readers by total body movement learning games.
(1986). The effects of coaching on the standardized test
performance of learning disabled and behaviorally disordered (Doctoral dissertation, University of Tulsa, 1973). Disserta-
tion Abstracts International, 34(7-A), 4049.
students. Remedial and Special Education, 7(5), 37-41.
*89. White, C. V., Pascarella, E. T., & Pflaum, S. W. (1981).
*73. Scruggs, T. E., & Tolfa, D. (1985). Improving the test-
Effects of training in sentence construction on the compre-
taking skills of learning-disabled students. Perceptual and hension of learning disabled children. Journal of Educational
Motor Skills, 60, 847-850.
Psychology, 71(5), 697-704.
*74. Simmonds, E. P. M. (1992). The effects of teacher *90. Williams, J. P., Brown, L. G., Silverstein, A. K., &
training and implementation of two methods for improving deCani, J. S. (1994). An instructional program in compre-
the comprehension skills of students with learning disabili- hension of narrative themes for adolescents with learning
ties. Learning Disabilities Research & Practice, 7, 194-198. disabilities. Learning Disability Quarterly, 17, 205-221.
*
75. Sinatra, R. C., Stahl-Gemake, J., & Berg, D. N. (1984). **91. Wise, B. W., Ring, J., Sessions, L., & Olson, R. K.
Improving reading comprehension of disabled readers through (1997). Phonological awareness with and without articula-
semantic mapping. The Reading Teacher, 38(1 ), 22-29. tion : A preliminary study. Learning Disability Quarterly.
*76. Smith, M. A. (1989). The efficacy of mnemonics for *92. Wong, B. Y. L., & Jones, W. (1982). Increasing meta-
teaching recognition of letter clusters to reading disabled comprehension in learning disabled and normally achieving
students. Dissertation Abstracts International, 50(5-A), 1259- students through self-questioning training. Learning Disabil-
1260. ity Quarterly, 5(3), 228-240