ArticlePDF Available

Phonemic Awareness Instruction Helps Children Learn to Read: Evidence From the National Reading Panel's Meta-Analysis

Authors:

Abstract and Figures

A quantitative meta-analysis evaluating the effects of phonemic awareness (PA) instruction on learning to read and spell was conducted by the National Reading Panel. There were 52 studies published in peer-reviewed journals, and these contributed 96 cases comparing the outcomes of treatment and control groups. Analysis of effect sizes revealed that the impact of PA instruction on helping children acquire PA was large and statistically significant (d = 0.86). PA instruction exerted a moderate, statistically significant impact oil reading (d = 0.53) and spelling (d = 0.59). Not only word reading but also reading comprehension benefited. PA instruction impacted reading under all the conditions examined although effect sizes were larger under some conditions. PA instruction helped various types of children: normally developing readers as well as at-risk and disabled readers: preschoolers, kindergartners, and first graders: low socioeconomic status children as well as mid-high SES. PA instruction improved reading, but it did not improve spelling in disabled readers. PA instruction was more effective when it was taught with letters than without letters. when one or two PA skills were taught than Multiple PA skills, when children were taught in small groups than individually or in classrooms. and when instruction lasted between 5 and 18 hours rather than longer, Classroom teachers were effective in Leaching PA to their Students. Effect sizes were larger for Studies using more rigorous experimental designs, with rigor assessments drawn from Troia ( 1999), In Sum, PA instruction was found to make a statistically significant contribution to reading acquisition.
Content may be subject to copyright.
Phonemic Awareness Instruction Helps Children Learn to Read: Evidence from the National
Reading Panel's Meta-Analysis
Author(s): Linnea C. Ehri, Simone R. Nunes, Dale M. Willows, Barbara Valeska Schuster,
Zohreh Yaghoub-Zadeh and Timothy Shanahan
Reviewed work(s):
Source:
Reading Research Quarterly,
Vol. 36, No. 3 (Jul. - Aug. - Sep., 2001), pp. 250-287
Published by: International Reading Association
Stable URL: http://www.jstor.org/stable/748111 .
Accessed: 14/08/2012 13:56
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
International Reading Association is collaborating with JSTOR to digitize, preserve and extend access to
Reading Research Quarterly.
http://www.jstor.org
Reading Research Quarterly
Vol. 36, No. 3
July/August/September 2001
?2001 International Reading Association
(pp. 250-287)
Phonemic
awareness
instruction
helps
children
learn
to
read:
Evidence
from
the
National
Reading
Panel's
meta-analysis
Linnea C. Ehri
City University
of
New York
Qraduate
Center,
New York,
USA
Simone R.
Nunes
Funda~go
Coordena~go
de
Aperfei?oamento
de Pessoal de
Nfvel
Superior, Ministry
of
Education, Brasilia,
Brazil
Dale
M. Willows
Barbara
Valeska Schuster
Zohreh
Yaghoub-Zadeh
Ontario
Institute
for
Studies
in
Education, Toronto,
Canada
Timothy
Shanahan
University
of
Illinois,
Chicago,
USA
n 1997,
the U.S.
Congress
directed
that a national
panel be convened to assess the status of research-
based knowledge regarding the effectiveness of vari-
ous approaches for teaching children to read. The
Director of the National Institute
of Child Health and
Human Development in consultation with the Secretary
of
Education constituted the National Reading Panel (NRP),
which was composed of 14 individuals. Members of the
(NRP) Panel formed subgroups to locate research examin-
ing various categories of studies. One of the subgroups
focused on alphabetics and conducted two meta-analyses,
one on phonemic awareness instruction
and one on
phonics instruction.
The final report was submitted to
Congress in April 2000. Results of the phonemic aware-
ness analysis are presented here.
When today's educators consider the ingredients of
effective programs to teach children to read, phonemic
awareness (abbreviated PA) receives much attention. The
purpose of this study was to examine the scientific evi-
dence supporting claims about phonemic awareness in-
struction. We sought answers to several questions: Is
phonemic awareness instruction effective in helping chil-
dren learn to read? Under what circumstances and for
which children is it most effective?
Were studies showing
its effectiveness designed to yield scientifically valid find-
ings? How applicable are these findings to classroom
practice?
There were several reasons why the NRP
selected
phonemic awareness instruction
for review and analysis.
First,
correlational
studies have identified phonemic aware-
250
ABTRCT
Phonemic awareness instruction helps children learn to read: Evidence from the National Reading
Panel's meta-analysis
A quantitative
meta-analysis evaluating
the effects
of phonemic
awareness
(PA)
instruction
on learning
to read
and
spell
was
con-
ducted
by the National
Reading
Panel. There
were
52 studies
pub-
lished
in peer-reviewed
journals,
and these
contributed
96 cases
comparing
the
outcomes
of treatment and control
groups. Analysis
of effect sizes revealed that the
impact
of PA
instruction
on helping
children
acquire
PA was
large
and
statistically significant
(d = 0.86).
PA
instruction exerted a moderate,
statistically
significant
impact
on
reading
(d = 0.53)
and
spelling
(d= 0.59).
Not
only
word
reading
but
also reading
comprehension
benefited.
PA
instruction
impacted
reading
tnder
all the conditions examined
although
effect sizes were
larger
tnder
some conditions. PA instruction
helped
various
types
of children:
normally developing
readers as well as at-risk and
disabled readers;
preschoolers, kindergartners,
and first
graders:
low
socioeconomic
status
children
as well
as mid-high
SES.
PA
instruc-
tion
improved reading,
but
it did not
improve spelling
in disabled
readers.
PA instruction
was more effective
when
it was
taught
with
letters than
without
letters,
when one or two
PA skills were
taught
than
multiple
PA
skills,
when
children were
taught
in small
groups
than
individually
or in classrooms,
and when
instruction
lasted
be-
tween
5 and 18
hours
rather than
longer.
Classroom teachers
were
effective
in teaching
PA to their
students.
Effect
sizes were
larger
for
studies
using
more
rigorous
experimental
designs,
with
rigor
as-
sessments
drawn from
Troia
(1999).
In sum,
PA
instruction
was
found
to make
a statistically significant
contribution
to reading
acquisition.
La Instrucci6n en Conciencia Fonemica Ayuda a los Ni-os a Aprender a Leer:
Evidencia del Meta-analisis del Panel Nacional de Lectura
Un
meta-anilisis cuantitativo
que
evalu6
los
efectos de la
instrucci6n
en conciencia
fonemica
(CF)
sobre
el aprendizaje
de la lectura
y la
escritura fue Ilevado a cabo por el Panel
Nacional de Lectura
(National
Reading
Panel).
Se tomaron
52 estudios,
publicados
en
revistas con referato,
que aportaron
96 casos
en los que
se com-
par6
el resultado de los
grupos
de tratamiento
y de control. El
anali-
sis de las
magnitudes
del efecto revel6
que
el impacto
de la instruc-
ci6n en CF
sobre
el desarrollo de la CF en los niiios
fue
grande
y
estadisticamente
significativo
(d = 0.86).
La
instrucci6n
en CF
ejer-
ci6
un
impacto
moderado,
estadisticamente
significativo
en la lectura
(d = 0.53)
y escritura
(d = 0.59).
No s6lo
se benefici6
la lectura
de
palabras,
sino
tambien
la comprensi6n
lectora. La
instrucci6n
en CF
impact6
en la lectura,
en todas las condiciones
examinadas,
sin
em-
bargo
las
magnitudes
del efecto
fueron
mayores
en algunas
condi-
ciones. La
instrucci6n
en
CF
ayud6
a distintos
tipos
de nihios:
lectores
de desarrollo
normal,
lectores de riesgo y lectores
con
dificultades;
niflos
de jardin, preescolar
y primer grado;
niifos
de NSE
bajo
y
nifios
de NSE
medio-alto. La
instrucci6n en CF
mejor6
la lectura
pero
no la
escritura
en los lectores
con
dificultades.
La
instruccion
en
CF
fue
mis efectiva
cuando
se realiz6 con apoyo
de las
letras
que
cuando
no se enseharon las
letras,
cuando
se enseflaron
una
o dos
habilidades
de CF
que
cuando
se enseflaron mflttiples
habilidades,
cuando
se insruy6
a los niflos
en pequefros
grupos que
cuando
se
realiz6
individualmente
o en el aula
y cuando
la
instrucci6n
dur6
en-
tre
5
y 18
horas
que
cuando
abarc6f
mIs
tiempo.
Los
docentes
fueron
eficaces en la
ensefianza
de la CF a los estudiantes.
Las
magnitudes
del efecto
fueron
mayores
en los estudios
que utilizaron
disefios
experimentales
mis rigurosos,
con
evaluaciones
de rigor
tomadas
de
Troia
(1999).
En
suma,
se hall6
que
la instrucci6n
en CF
constituye
una
contribuci6n
significativa
a la
adquisicion
de la lectura.
Phonemische BewuBtseinsanweisung hilft Kindern
beim Erlernen
des Lesens: Beweise aus der
Meta-Analyse des Nationalen Lesegremiums
Eine
quantitative
Meta-Analyse,
welche
Auswirkungen
phonischen
Aufnahmebewugseins
(PA)
beim
Erlernen des Lesens
und
Buch-
stabierens
bewertet,
wurde
vom Nationalen
Lesegremium
durchge-
f'hrt.
Es
wurden
52 Studien
in von Lehrern
rezensierten Fachzeit-
schriften
publiziert
und diese nannten
96 Fille, bei denen die
Ergebnisse
von
Anwendungs-
und
Kontrollgruppen
verglichen
wur-
den.
Die
Analyse
iber
WirksamkeitsausmaSe
ergab,
daB
der
Einflug
von
PA-Anweisungen
in der
Unterst0tzung
der
Kinder
beim
Erwerb
von PA grog und statistisch
bedeutend
(d = 0.86) war. PA-
Anweisungen
iibten
einen
ausgleichenden,
statitisch
bedeutsamen
Einflug
aufs
Lesen
(d = 0.53)
und
Buchstabieren
(d = 0.59)
aus.
Nicht
nur das
Lesen
von
Wbrtern,
sondern
auch
das
Leseverstiindnis
wurde
begdnstigt.
PA-Anweisungen
beeinfluSten
das
Lesen
unter
all
den
untersuchten
Bedingungen,
obwohl
die
Ausmaf
e der
Aus-
wirkungen
bei
einigen Bedingungen
gr6ger
waren.
PA-Anweisungen
halfen
unterschiedlich
gearteten
Kindern:
normal
sich entwickelnde
Leser,
sowie
auch
risikobedingte
und
behinderte
Leser;
Vorschdler,
Kindergartenteilnehmer
und
Erstklissler;
Kinder der
unteren
Gund-
schulstufen
(SES)
als auch mittlerer
und
h6herer
Grundschulstufen
(SES).
PA-Anweisungen
verbesserten
das
Lesen,
nicht
jedoch
das
Buchstabieren
bei
behinderten
Lesern.
PA-Anweisungen
waren
wirk-
samer,
wenn sie zusammen
mit
den Buchstaben
statt
ohne
Buch-
staben
unterrichtet
wurden,
wenn
jeweils
eine oder zwei
PA-Fertig-
keiten statt
ein Mehrfaches an
PA-Fertigkeiten
unterrichtet
wurde,
wenn Kinder in kleinen
Gruppen
statt einzeln
unterrichtet
oder
in
Klassenriumen,
und
wenn
der
Unterricht
insgesamt
zwischen
5 und
18
Stunden
andauerte,
statt
langfristiger.
Klassenlehrer
waren
beim
Unterricht
von PA
erfolgreich
mit ihren
Schillern.
Das
Wirkungsaus-
maS war
unter
Nutzung
strengerer
Experimentierpline
ffor den
Unterricht
grb6er,
mit
den von Troia
(1999)
ahgeleiteten
straffen
Bewertungen. Zusammengefa&t
fand
man.
dag
die
PA-Anweisung
einen statistisch
nachweisbar
bedeutenden
Beitrag
ztm Erwerb
von
Lesefertigkeiten
leisten.
251
ABTAT
if stiff: br k b lra c i,. - ?r efib :
f -j -f ]L ? 1] - 3/ 3B Cr t: b a c;a ?3
L I:
(PT PA) 41)-77 1)
f iZRA t
t tft52 ?
OFf 3 %'9 .L
t L ;It
#7 -T-j -f ;<
P )A -#r
%IZ
10 , 7-A 7V
PA ?
-
A
1:4f-4t
Z PAA l
?')-7-f'l:Zt' t =JJZ
cA
<
*
it
A?,
4il,
IC?
J5
Z
(d-
=
0.
86)
1"'L
7)
T
49
;
7J-
1:-
45-
4:
o
PAr~l
4
It
1
-
-7 -f
:/
Y
IZ
r?
f
L,
Clls~
0.53), A ')
v1); Y ZC
, L,
-C : d = 0.59)~
L
VA A
rtJt 5
ZA40K C
L
-C?)Z <tc07Dr-
0) )
PA A`
t4-cQ);
It IPA
It
7'; l
t4% oc PA 1
J: I 4C
3'
-
f
14t M, -102
1~~'{92?stc0 PA t?i
P IJ
;?
-Lt
,
h?
-
-D
tz
o
PA
11
?t
L,
-C;i~-LS
t
Z Ut ) c T 1
t;c0
PA :IUAk,14
11 -Akt
PAQ0)
A A/
1 1 7)-
1 2
--2)0)PAQ0)A A-+r
)1
~~~YIL-~P
T~~i~~t5
*Mhl,;?
18
*FilioR
Q)-A~: PA Y *i*ttct ?
fMlt
19 Mrk,
(-,
Z 19f53) -C{IIttct
x PA<
4-c
W'k-15ifhtn~
Troia
(1999)7),-;14;
htz
og
tZ
3
:
PA
L'enseignement de la conscience phonemique aide les enfants a apprendre a lire : Preuves
provenant de la meta-analyse de la Commission Nationale de Lecture
Une
meta-analyse quantitative
pour
evaluer les
effets de lenseigne-
ment de la conscience
phonemique
(CP)
sur
l'apprentissage
de la
lecture et de
l'ecriture
a
ete conduite
par
la Commission
Nationale de
Lecture. Elle a porte
sur
52
etudes
publiees
dans
des journaux
;
comite de redaction
ayant
conduit
r 96 etudes
comparant
les resul-
tats de groupes experimentaux
et contr1les.
L'analyse
de l'impor-
tance des effets
revele
que
1'enseignement
de la CP
sur
le developpe-
ment de la CP a un impact large
et statistiquement significatif
(d =
0.86).
L'enseignement
de la CP
a un
effet
modere,
statistiquement
significatif
(d = 0.53) sur
l'criture
(d = 0.59).
Non
seulement
la lec-
ture de mots en beneficie mais
6galement
la comprehension.
L'enseignement
de la CP
a un
impact
sur
la lecture dans
toutes
les
conditions
examinees,
bien
que
la taille
de leffet
soit
plus
large
dans
certaines
conditions.
L'enseignement
de la CP
est
une
aide
pour
dif-
firents
types
d'enfants
: aussi
bien
pour
ceux
qui apprennent
a lire
normalement
que pour
les enfants A
risque
et pour
ceux
qui
ont
des difficultes
; pour
les enfants
d'ecole
maternelle ou de premiere
annee
de primaire
: pour
les enfants
de milieu
social defavorise
aus-
si bien
que
pour
ceux de classe
moyenne.
L'enseignement
de la CP
facilite
le developpement
de la lecture
mais
pas
celui
de l'Ocriture
chez les enfants
en difficult&.
L'enseignement
de la CP a plus
d'effet
quand
il a lieu
avec des lettres
que
sans
lettres,
quand
on enseigne
une
ou deux
competences
de CP
que
de multiples competences
de
CP,
quand
il a lieu en petit
groupe
que
individuellement
ou
en salle
de classe,
et quand I'enseignement
dure
entre
5 et 18
heures
plut6t
que
plus
longtemps.
L'enseignement
de la CP
en classe
par
des
en-
seignants
est efficace.
La taille des
effets
est plus
grande
quand
les
etudes
utilisent
des
plans
experimentaux
plus
rigoureux,
et des Mval-
uations
rigoureuses
provenant
de Troia
(1999).
En
bref,
il apparait
que
l'enseignement
de la CP
apporte
une
contribution
statistique-
ment
significative
a l'acquisition
de la lecture.
0DOHeMHqecKoe o6yqeHHe noMoraeT eTTHM
yqHTbC$I SIHTaTL:
4HTorH MeTa-aHaJIH3a, nponeneHHOro HauHOHaJIbHbIM COBeTOM TeHHSI
HaHOHarJbHbIfl COBeT 4TeHHA npeanpHHRn KOJIHeCCTBeHHbIl4
MeTa-aHaJIH3 34CeKTHBHOCTH OHeMH'eCKoro ("3ByKO-
6yKBeHHoro") MeToaa (PA) npH o6yqeHHHq TeHHIO H
opporpanHH. B HayqHbIX KypHHanaX
ony6nHKOBaHbl
pe3ynJbTaTb 52 HccneIOBaHHar, r•C OnHcaHbl 96 cnyqaeB
cpaBHeHHuS
OfbITHO H KOHTpOJIbHOr rpynn.
KpynHoMacmTa6HbI aJanIH3 BbIAlBH, HTO
poJnb PA-o6yqeHHSi
B o6peTeHHH PA-HaBbIKOB BCJeHKa
H CTaTHCTHqeCKH
3HaqHMa
(d = 0.86). PA-o6yqeHHe OKa3blBaeT cpe•Hee, CTaTHCTHeCCKH
3HaIHMOe BJIH•MHHe
Ha qTeCHe (d = 0.53) H opcorpa•ino (d =
0.59). PA-o6yqeHHe lHOJe3HO
He TOJIbKO AUJI
'TeHHAS
OTLJenjbHbIX
CJIOB,
HO H AjJISl OHHMaHHS HpO'qHTaHHOFO
TeKcTa.
PA-o6yseHHe BJnIHeT
Ha iTeCHe BO Bcex cJnyqax, XOTI npH
HeKOTOPbIX
YCJIOBHSX
34jeKT OKa3aJICCI 6OnJbHM. PA-
o6yqeHHe noMoraeT pa3HbIM THHaM
AJeTer: HOpManJbHO
pa3BHBraloIUHMCs,
npea•TaBHTeJIsM rpyHHbl pHCKa H AeTAM
C
orpaHHNeHHbIMH BO3MO)KHOCTSMH;I
nepBOKJIacCHHKaM
H
AOUIKORbHHKaM,
nOCeCIOLUmIHM
H H HnOCeCaIOLuHM
A•TCKHA
caA; AeTrM C HH3KHM,
cpeIHHM H BblCOKHM
COUHaJIbHO-
3KOHOMHIeCKHM
CTaTyCOM. CpeaH npo6neMHblX
AeTerl
PA-
o6yqeHHe cnoco6CTByeT
ynymJeHHIO
HaBbIKOB
MTeHHS1,
HO
He opcorpampH. (OHeMHqCCKOC
O6yqeHHe
6onee
344eKTHBHO: B coqeTaHHH C 6yKBeHHbIM, HC)KenH
6e3 Hero;
fpH pa3BHTHH
OAHOrO-AIByX,
He)KenH MHOFO'HCJIeHHbIX
HaBbIKOB;
npH pa6oTe B
MaibIX
rpynnax, B oTJIHqie
oT
HHtAHBHayanbHOrO HJIH KnaCCHOro
o6ySeHHA; npH
npoIonJIKHTejIbHOCTH Kypca He 6onee 5-18 4acOB.
IUKOJIbHbie
y'HTenJ
c PA-o6yqeHHeM
cnpanBJIOTC5
ycnefHO.
IpH
Hcnonb3OBaHHH
6onee
CTporFx
3KcnepHMeHTaaJbHbIX
pa3pa6OTOK
C )KCCTKOfr
CHCTeMOg
OUeHKH,
pa3pa60TaHHOr4
Tpoia
(1999),
3a4eKTrHBHOCTb
Bblue.
B CJenoM
npH3HaHO,
4TO
PA-o6yreHHe
BHOCHT
CTaTHCTHeCCKH
3Ha4HMblri BKJIaa
B pa3BHTHe HaBbIKOB
TeHHMR.
252
ness and letter knowledge as the two best school-entry
predictors of how well children will learn to read during
the first
2 years of instruction
(Share,
Jorm, Maclean, &
Matthews, 1984). Second, many experimental studies have
been conducted to evaluate the effectiveness of PA instruc-
tion in facilitating reading acquisition. However, a recent
critique appearing in Reading Research
Quarterly suggested
that some of the PA studies are seriously flawed (Troia,
1999). Third,
there is currently
much interest in PA pro-
grams among teachers, principals,
publishers, and even
legislators
because they are claimed to be effective in im-
proving children's
success in learning to read.
To evaluate the adequacy and strength of the evi-
dence, we conducted a meta-analysis. We located all the
experimental studies that administered PA instruction to
students, that included a control group receiving non-PA
instruction or no special instruction, and that measured
the impact of PA instruction on reading outcomes. We
found 52 published studies that met our criteria. The
studies varied in many respects. Different types of phone-
mic awareness were taught. The participants
ranged from
preschoolers to sixth graders and included students at
risk for reading problems as well as students classified as
reading disabled. The instruction was delivered by class-
room teachers in some studies and by researchers or
computers in other studies. Children were tutored indi-
vidually, or they received instruction in small groups or
larger classroom groups. The statistic
we used was effect
size, that is, the difference between mean performance of
the treatment and control groups in standard deviation
units. The meta-analytic procedure allowed us to exam-
ine not only whether PA instruction exerted a statistically
significant impact on reading across all of these variations
but also whether these variations influenced the size of
the impact.
What is phonemic awareness? Phonemes are the
smallest units comprising spoken language. English con-
sists of about 41 phonemes. Phonemes combine to form
syllables and words. A few words have only one
phoneme, such as a or ob. Most words consist of a blend
of phonemes, such as go with 2 phonemes, or check with
3 phonemes, or stop with 4 phonemes. In the text below,
individual phonemes are represented with IPA (Interna-
tional Phonetic Alphabet) symbols between backslashes
(e.g., /g/) to contrast them with letters represented by
capitals (e.g., G). Phonemes are different from graphemes,
which are units of written language and represent
phonemes in the spellings of words (Venezky, 1970,
1999). Graphemes may consist of one letter, for example,
P, T, K, A, N; or multiple letters, CH, SH, TH, -CK, EA, -
IGH, each symbolizing one phoneme.
Phonemic awareness refers to the ability to focus on
and manipulate phonemes in spoken words (Liberman,
Shankweiler, Fischer, & Carter,
1974). Simply discriminat-
ing phonemes in words-for example, recognizing that
tan sounds different from Dan-is not PA. PA is different
from phonological awareness, which is a more encom-
passing term referring
not only to PA but also to aware-
ness of larger spoken units such as syllables and rhyming
words. Although PA studies often begin by teaching chil-
dren to analyze larger units, we did not include such
studies in our data set unless the instruction progressed
to the phonemic level.
Researchers have used the following tasks to assess
children's PA and to improve their PA through instruction
and practice:
1. Phoneme isolation, which requires recognizing
individual sounds in words; for example, "Tell
me the first sound in paste." (/p/)
2. Phoneme identity, which requires recognizing
the common sound in different words; for exam-
ple, "Tell me the sound that is the same in bike,
boy, and bell."
(/b/)
3. Phoneme categorization, which requires recog-
nizing the word with the odd sound in a se-
quence of three or four words; for example,
"Which word does not belong? bus, bun, rug."
(rug)
4. Phoneme blending, which requires listening to a
sequence of separately spoken sounds and com-
bining them to form a recognizable word; for ex-
ample, "What
word is /s/ /k/ /u/ /1/?"
(school)
5. Phoneme segmentation, which requires breaking
a word into its sounds by tapping out or count-
ing the sounds or by pronouncing and position-
ing a marker for each sound; for example, "How
many phonemes in ship?"
(3: /s/ /i/ /p/)
6. Phoneme deletion, which requires recognizing
what word remains when a specified phoneme is
removed; for example, "What is smile without the
/s/?" (mile)
In a few studies, instruction was focused on teach-
ing children to manipulate onsets and rimes in words
(Fox & Routh, 1984; Lovett, Barron, Forbes, Cuksts, &
Steinbach, 1994; Treiman & Baron, 1983; Wilson &
Frederickson, 1995). The onset is the single consonant or
blend that precedes the vowel, and the rime is the vowel
and following consonants; for example, j-ump, st-op,
str-ong. We included these studies because students were
essentially manipulating phonemes when the onset was a
single phoneme.
Some forms of PA instruction would qualify as
phonics instruction, which involves teaching students
Meta-analysis of phonemic awareness instruction 253
how to use grapheme-phoneme
correspondences
to
decode or spell words. What distinguished PA studies
from the general pool of phonics studies, however, is that
instruction given to students in the treatment groups, but
withheld from controls, was limited to grapheme-
phoneme manipulation and did not go beyond this to in-
clude other activities such as reading decodable text or
writing stories.
Contribution of PA in learning to read. PA is one of
the best predictors of how well children will learn to
read. Share et al. (1984) assessed kindergartners
on many
measures when they entered school, including phonemic
segmentation, letter name knowledge, memory for sen-
tences, vocabulary, father's occupational status, parental
reports of reading to children, and TV watching. They ex-
amined which of these measures best predicted how well
the children would be reading at the end of kindergarten
and at the end of first grade. Results showed that phone-
mic awareness was the top predictor along with letter
knowledge. PA correlated r = 0.66 with reading achieve-
ment in kindergarten
and r = 0.62 in first grade.
PA is thought to contribute to helping children learn
to read because the structure
of the English writing sys-
tem is alphabetic. Moreover, it is not easy to figure out
the system. Words have prescribed spellings that consist
of graphemes symbolizing phonemes in predictable
ways. Being able to distinguish the separate phonemes in
pronunciations of words so that they can be matched up
to graphemes is difficult. There are no breaks in speech
signaling where one phoneme ends and the next one be-
gins. Rather,
phonemes are folded into one another and
are coarticulated. Discovering phonemic units is helped
greatly by explicit instruction in how the system works.
This is underscored by research revealing that people
who have not learned to read and write have great trou-
ble performing phonemic awareness tasks (Morais,
Bertelson, Cary, & Alegria, 1987). Likewise people who
have learned to read in a script that is not graphophone-
mic, such as Chinese, have difficulty segmenting speech
into phonemes (Mann, 1987; Read, Zhang, Nie, & Ding,
1987).
PA is thought to contribute to children's ability to
read words in various ways (Ehri, 1991, 1994). Decoding
words requires blending skill to transform
graphemes
into recognizable words. Reading words by analogy (e.g.,
reading brick by knowing how to read kick) requires
onset-rime segmentation and blending skill. Reading
words from memory by sight requires phoneme segmen-
tation skill. To store individual sight words in memory,
children need to match up graphemes to phonemes in
the word and retain these connections in memory (Ehri,
1980, 1992; Ehri & Wilce, 1987a; Rack, Hulme, Snowling,
& Wightman, 1994; Reitsma, 1983).
The impact of PA instruction on various types of
word reading was examined in many of the studies. The
simplest word reading task given to preschoolers re-
quired them to look at a word (sat) and decide whether
it said "sat"
or "mat"
(Byrne & Fielding-Barnsley, 1991).
Studies with older children assessed their ability to read a
list of words or their ability to remember how to read
words after practicing them. Pseudoword reading tasks
assessed children's ability to read novel words such as
feem, bote, and cliss.
PA is thought to help children write words in vari-
ous ways. They may invent letter-sound spellings or re-
trieve correct spellings from memory. Both processes
require phonemic segmentation skill, enabling spellers to
match up sounds to letters in words (Griffith,
1991). The
impact of PA instruction on spelling skill was assessed in
several studies.
Some of the studies measured reading comprehen-
sion. To comprehend a text, readers must be able to read
most of the words. However, other capabilities influence
reading comprehension as well, such as readers' vocabu-
lary, their world knowledge, and their memory for the
text. We expected PA instruction to benefit children's
reading comprehension because it depends on effective
word reading. However, we did not expect the effect to
be as strong on comprehension as on word reading be-
cause the influence is indirect.
Replication and extension of a previous meta-
analysis. The present study attempted to replicate and
extend findings of the meta-analysis of phonemic aware-
ness instruction studies published by Bus and van
IJzendoorn (1999), referred to here as the B&I
study.
Some of the same moderator variables were examined,
including grade, reading ability, teaching with or without
letters, instructor,
size of the group taught, and whether
the control group was treated or untreated. The present
study expanded the database that tested the effects of PA
instruction on reading outcomes from 34 studies to 52
studies. In the B&I analysis, when studies included more
than one PA treatment or more than one control group,
these groups were combined into single treatment-control
comparisons. In contrast, we kept these comparisons sep-
arate in our analysis. As a result, the number of treat-
ment-control comparisons increased from 34 in the B&I
study to 96 in our study. Having more comparisons made
it possible to expand the examination of moderator vari-
ables. Also spelling outcomes were evaluated in more
studies. Having Troia's (1999) critique of PA studies al-
lowed us to examine, more thoroughly than B&I
could,
the relationship between methodological design features
and the size of effects detected in the experiments.
Methodological rigor ofPA studies. Many correla-
tional studies have reported strong relationships between
254 READING
RESEARCH
QUARTERLY July/August/September 2001 36/3
phonemic
awareness
and learning
to read
(for reviews
see Blachman,
2000; Ehri,
1979;
Snow,
Burns,
& Griffin,
1998;
Stahl
& Murray,
1994;
Wagner
& Torgesen,
1987).
However,
correlational
findings
are insufficient
to show
that
PA was the underlying
cause enabling
some students
to read
better
than
others.
This is because correlational
findings
do not rule
out other
factors that
might
also ex-
plain
the relationship.
To show that
PA
operates
as a di-
rect
cause in helping
children
learn
to read,
evidence
from
experimental
studies
with treatment
and control
groups
is needed.
Well-designed
experiments
yielding
positive
out-
comes provide
the strongest
evidence
that
PA caused
the
improvement
in reading.
Although
all of the studies
in
our database consisted
of experiments,
some were better
designed
than others.
Studies
varied
in whether
they used
treated
or untreated control
groups.
The use of untreated
controls
receiving
no special
attention
from researchers
runs
the risk of Hawthorne
effects
as an explanation
for
differences
favoring
the treatment
group.
Studies
varied
in whether
students
were randomly
assigned
to treatment
and control
groups,
or whether a quasi-experimental
de-
sign was used in which existing groups
were assigned
to
conditions,
or whether students
were matched
and as-
signed to conditions.
Although
random
assignment
is
preferable,
researchers
may be limited
to a quasi-experi-
mental
design when classrooms
in schools are studied.
We examined
whether
positive
effects
of PA instruction
emerged primarily
from the weaker
designs
or whether
effects were strongest
in the best designed experiments.
Surprisingly,
B&I
(1999)
found
larger
effect
sizes for
comparisons
that
contrasted
PA-instructed
groups
to treated
control
groups
than to untreated
control
groups.
This
finding
is the opposite
of what
would be expected
if
Hawthorne effects
were inflating
effect
sizes. Whether this
finding replicates
in our database became
a question
of
special
interest.
In a recent
critique
of PA
instruction
studies,
Troia
(1999) identified several
design flaws and applied
these
criteria
to rate the studies
for their
methodological rigor.
He considered
threats
to internal
validity
(i.e., attribution
of cause to the treatment)
as well as to external
validity
(i.e., generalizability
of findings).
To evaluate whether
these flaws
might
be associated with outcomes,
we ex-
amined the relationship
between Troia's
ratings
and the
effect sizes in the studies. Our purpose was to determine
whether claims about PA instructional effects are supported
mainly by poorly-designed or well-designed studies.
Other issues. One issue of interest was whether PA
instruction might be more effective for some age/grade or
reader groups than for others. Preschoolers use language
as a tool for communication, so their focus is upon the
meaning of speech, not upon its phonological structure.
As a result they have little PA and hence stand to gain
much from PA instruction. In contrast, beginning readers
have acquired at least some awareness of phonemes,
even without PA instruction, because making progress in
reading requires grapheme-phoneme knowledge. PA in-
struction may still contribute to their growth in literacy,
but its impact may be less than in the earliest grades. As
readers and writers advance beyond first grade, the need
to acquire additional PA may become less important than
the need to learn about spelling patterns in words, so in-
struction focused on phonemes may yield diminishing re-
turns. This suggests that the impact of PA instruction may
be greatest in preschool and kindergarten, and may be-
come smaller beyond first grade. In the B&I (1999) meta-
analysis, although all groups profited from PA instruction,
preschoolers benefited more than kindergartners
or pri-
mary school students.
PA instruction
may contribute less to older, normally
developing readers, but it may make a big difference for
older children who have failed to make normal progress
in learning to read. Research has shown that disabled
readers have poor phonemic awareness, even below that
of nondisabled students reading at the same grade-equiv-
alent level (Bradley & Bryant, 1983; Bruck, 1992; Fawcett
& Nicholson, 1995). In addition, disabled readers have
special difficulty learning to spell (Bruck, 1993). We
might expect PA instruction to help in remediating the
reading and spelling difficulties of these readers.
Instruction in PA may be conducted with or without
alphabetic letters. In some studies, children were taught
to manipulate phonemes in words by using letters as
markers for the sounds whereas in other studies children
were taught to work with spoken units only. Sounds are
ephemeral, short-lived, and hard to grasp, whereas letters
provide concrete, visible symbols for phonemes. Thus,
we might expect children to have an easier time acquir-
ing PA when they are given letters to manipulate. Also,
because letters bring children closer to the task of apply-
ing PA in reading and spelling, we would expect transfer
to be greater when PA is taught with letters. In the B&I
(1999) study, PA instruction
with letters produced larger
effects on PA and reading than instruction without letters.
English includes many words that exhibit grapheme-
phoneme regularity,
but it also includes words that derive
their regularity
from larger,
word-based spelling patterns
or that contain exceptions to graphophonemic regularities.
This contrasts with other written languages whose
spellings are regular principally at the grapheme-phoneme
level. The studies in our database provided PA instruction
not only in English but also in other languages, namely,
Norwegian, Finnish, Swedish, Danish, Spanish, Hebrew,
Dutch, and German. In most of these languages, the
grapheme-phoneme connections are more transparent
Meta-analysis of phonemic awareness instruction 255
than they are in English. Of interest was whether PA in-
struction might contribute more to reading acquisition in
English because children need more help in figuring out a
nontransparent
graphophonemic system.
Because classroom teachers are the purveyors of
reading instruction for most children, it is important to
verify that they can teach PA effectively. Some of the
studies used classroom teachers who taught PA to their
students, thus allowing us to assess their effectiveness.
There is substantial evidence that 1:1 tutoring is the
most effective form of instruction (Bloom, 1984; Cohen,
Kulik, & Kulik, 1982; Glass, Cahen, Smith, & Filby, 1982;
Pinnell, Lyons, DeFord, Bryk, & Seltzer, 1994; Wasik &
Slavin, 1993). It allows instructors to tailor lessons to stu-
dents' needs and difficulties. In the database, PA instruc-
tion was delivered not only to individual children but
also to small groups and to classrooms of students. We
expected tutoring to produce stronger effects than the
other two types of delivery. However, B&I (1999) ob-
tained only partial support for this. They found that indi-
vidualized instruction
was less effective than small-group
instruction for teaching PA, but was more effective for
promoting transfer
to reading. Replication of this effect
with our larger database was considered important.
Method
Database. We conducted an electronic search of two
databases, ERIC
and PsychInfo. Six terms involving
phonemic awareness (PA) were crossed with 15 terms re-
lated to reading performance. The PA terms were phone-
mic awareness, phonological awareness, spelling,
blending, learning to spell, and invented spelling. The
reading terms were reading, reading ability, reading
achievement, reading comprehension, reading develop-
ment, reading disabilities, reading skills, remedial reading,
beginning reading, beginning reading instruction, read-
ing acquisition, word identification, word reading, oral
reading, and miscues. Only articles appearing in journals
written in English were considered. The search yielded
637 articles through ERIC and 1,325 articles through
PsychInfo. In addition, we hand-searched references in
the studies screened and in several review papers
(Apthorp, 1998; Blachman, 2000; Bus & van IJzendoorn,
1999; Stahl
& Murray,
1994; Troia, 1999; Wagner, 1988).
To qualify for our analysis, studies had to meet the
following criteria:
1. Studies had to adopt an experimental or quasi-
experimental design with a control group.
2. Studies had to appear in a refereed journal.
3. Studies had to test the hypothesis that instruction
in phonemic awareness improves reading
performance over alternative
forms of instruction
or no instruction.
4. Studies had to provide instruction in phonemic
awareness that was not confounded with other
instructional
methods or activities.
5. Studies had to report statistics permitting the cal-
culation or estimation of effect sizes.
The requirement that studies come from refereed
journals was adopted by the National Reading Panel and
applied to all the meta-analyses. Our rationale was that
limiting studies to those passing the test of peer review
serves to minimize the risk of admitting studies of poor
quality.
The final set of studies meeting our criteria
num-
bered 52. From these, 96 cases comparing individual
treatment and control groups were derived. Because
some of the studies included multiple treatment or con-
trol groups, our cases included comparisons utilizing the
same group more than once. Seven treatment
groups
were included twice because they were compared to two
different control groups. Sixteen control groups were in-
cluded twice because they were compared to two differ-
ent treatment groups. One control group appeared three
times because it was compared to three treatment
groups.
Although this meant that effect sizes were not completely
independent across cases, we preferred this alternative
to
that of combining treatment and control groups within
studies because we did not want to obscure important
moderator variables of interest.
The following studies included treatment or control
groups that were not deemed appropriate for our analy-
sis, so these comparisons were not included: a treatment
group given decoding instruction and word reading
(Barker
& Torgesen, 1995); a treatment
group given a
reading and writing program (Brennan & Ireson, 1997); a
treatment group taught to manipulate syllables rather
than phonemes (Sanchez & Rueda, 1991); a treatment
group taught semantic categorization with written words
(Defior & Tudela, 1994); treatment
groups in which the
instructors failed to spend the time prescribed for training
(Olofsson & Lundberg, 1983); treatment groups in which
children not only analyzed phonemes but also read
words in sentences and stories, unlike children in the
control groups who only listened to stories or remained
in their classrooms (Solity, 1996; Weiner, 1994); a control
group lacking not only PA instruction
but also the
Reading Recovery instruction given the treatment
group
(Iversen & Tunmer, 1993); a control group that did not
control for all of the non-PA elements of instruction
(Lovett et al., 1994; Vellutino & Scanlon, 1987).
The studies in our database were coded for several
characteristics. These are listed in Table 1. Three out-
256 READING RESEARCH QUARTERLY July/August/September 2001 36/3
comes were of primary
interest: phonemic awareness,
reading, and spelling. Some studies measured one or an-
other of the three outcomes in multiple ways and reported
means and standard deviations on each measure; for ex-
ample, measuring children's ability to segment words into
phonemes, to delete phonemes from words, and to blend
phonemes into words. Because we wanted one overall
effect size for each outcome from each study, we com-
bined measures of the same outcome within each study
by first calculating effect sizes on each measure and then
averaging effect sizes across measures to create one effect
size for the outcome of interest. The combined measure
for phonemic awareness was limited to tasks assessing
phoneme manipulation, not larger unit manipulation. The
kinds of measures that were combined into the reading
outcome included word reading, pseudoword reading,
reading comprehension, oral text reading, reading speed,
time to reach a criterion of learning, and miscues. The
spelling composite included measures of invented
spellings as well as correct spellings of words and pseu-
dowords. In this way, each treatment-control
group com-
parison had the potential for contributing one effect size
to the pool of effect sizes for phonemic awareness, for
reading, and for spelling outcomes.
In addition, we were interested in the effect of PA
instruction on specific measures of PA (i.e., segmentation,
blending, deletion, and other), on specific measures of
reading (i.e., words, nonwords, comprehension), and on
outcomes measured immediately after testing as well as
after a delay. When studies administered delayed
posttests at more than one point in time, we calculated
effect sizes for the first two test points. Separate effect
sizes on these measures were included for each study in
the database. Also we compared effect sizes on experi-
menter-devised and standardized tests of reading and
spelling. Standardized
tests are valued because they give
a more general picture of transfer effects. However, when
standardized tests are designed to distinguish individuals
across many ages or grades, they may be less sensitive to
differences within a narrow range of performance.
Because several studies examined whether PA in-
struction impacted students' performance in math, we in-
cluded this as well. It was expected that scores in math
would not show gains, hence ruling out Hawthorne
effects.
There were three reader
groups. One group consisted
of children who were not distinguished as having any
reading problems, referred to as normally developing
readers. One group consisted of children below second
grade who were at risk for developing reading difficul-
ties. They were given this label by authors of the studies.
Being at risk was indicated by low PA or low reading in
83% of the cases and low socioeconomic status in 27%
of
the cases. Other at-risk indicators were developmental or
language delays or cognitive disabilities.
The third reader group consisted of reading dis-
abled (RD) students who had already developed reading
problems. All but three cases involved children between
second and sixth grades. The three cases were first
graders who qualified for Reading Recovery programs
(Hatcher, Hulme, & Ellis, 1994; Iversen & Tunmer, 1993).
In most studies, being reading disabled meant reading
below grade level despite at least average cognitive ability.
In one study, the school's definition of learning disabled
was used (Williams, 1980). In one study, students had not
only RD but also neurological impairment and language-
learning problems (Lovett et al., 1994).
PA programs varied in whether they focused on
specific PA manipulations. Single-focus studies taught
blending, categorization, identity, segmentation, or onset-
rime. Double-focus studies involved combinations of
blending, segmenting, deletion, or categorization. Global
treatments taught three or more PA skills. Programs
that
taught onset-rime manipulation were coded as having a
single focus even though the instruction might have
taught children to blend or segment the onsets and rimes
(e.g., Fox & Routh, 1976). Instruction varied in whether
children were taught to manipulate phonemes using let-
ters or whether attention was limited to phonemes in
speech. Instruction that had children manipulate blank
markers was coded as a nonletter treatment.
The instructional
delivery unit varied across studies.
Students were tutored individually or taught in small
groups or whole classrooms. The size of the small groups
varied from two to seven students. The identity of in-
structors
varied across studies and included classroom
teachers, researchers or their assistants, or computers.
Credentialed teachers who conducted the instruction but
were not the students' classroom teacher were coded as
researchers. We coded studies to reflect whether fidelity
to treatment
was checked, that is, whether researchers
observed instructors
to assess their adherence to treat-
ment procedures.
Some features of the experimental methodology
were coded. One feature was group assignment. In some
studies, children were randomly assigned to conditions.
In some studies, children were members of intact groups
that were assigned to conditions, referred to as nonequiv-
alent groups. When two classrooms were assigned ran-
domly, one to the treatment and one to the control
condition, they were categorized as nonequivalent
groups. When several classrooms were assigned randomly
to treatment and control conditions, they were catego-
rized as random assignment. The third way of assigning
children to conditions involved matching children on the
basis of similar test scores. Although members of a match
Meta-analysis of phonemic awareness instruction 257
Table 1 Dependent and moderator variables included in the meta-analyses
Outcome measures
1. Composite measures: phonemic awareness; reading; spelling
2. Measures of phonemic awareness: segmentation; blending; deletion; other
3. Measures of reading:
Standardized
versus experimenter-devised tests of word reading
Standardized versus experimenter-devised tests of nonword reading
Reading comprehension
4. Measures of spelling: Standardized versus experimenter-devised tests of spelling
5. Measure of math achievement
6. Test points:
Immediately after instruction
First
follow-up test (delay of 2 to 15 months)
Second follow-up test (delay of 7 to 36 months)
Characteristics of participants
1. Reader level: at-risk readers; disabled readers; normally progressing readers
2. Grade level: preschool; kindergarten;
first grade; second through sixth grades
3. Socioeconomic status: low-SES;
middle-to-high-SES
4. Language:
English; other (Danish, Dutch, Finnish, German, Hebrew, Norwegian, Spanish, Swedish)
Properties of phonemic awareness instruction
1. PA skills taught:
a. single skill; two skills; three or more skills
b. segmenting and blending versus three or more skills
2. Use of letters: phonemes and letters manipulated versus only phonemes manipulated
3. Delivery unit: individuals; small groups (two to seven students); classrooms
4. Identity of instructor: classroom teachers; computers; researchers/others
5. Length of instruction:
ranged from 1 hour to 75 hours
Features of the design
1. Group assignment: random; matched; nonequivalent
2. Fidelity of instructors:
checked versus not checked or not reported
3. Control group: alternative
treatment;
no treatment
4. Size of the sample: ranged from 9 to 383 students
5. Internal
validity (from Troia, 1999):
Percentage of criteria
met
Number of critical flaws
6. External
validity (from Troia, 1999):
Percentage of criteria met
Number of critical flaws
7. Methodological rigor (from Troia, 1999):
Overall ranking
Characteristics of the study
Year of publication (1976 to 2000)
258 READING
RESEARCH
QUARTERLY July/August/September 2001 36/3
are typically assigned randomly to conditions, in some
studies this step was not stated explicitly.
To evaluate the relationship between the method-
ological quality of studies and the effect sizes found, we
adopted the five methodological criteria
applied by Troia
(1999) in his critique of the internal and external validity
of PA studies. Internal
validity refers to whether the treat-
ment rather
than some other variable caused the outcome
observed. External
validity refers to the generalizability of
the findings, that is, whether or not the results of a study
can be applied to other persons in other settings at other
times. To evaluate the internal and external validity of
studies, Troia used five summary measures: percentage of
internal validity criteria
met by the studies, number of
critical flaws challenging a study's internal validity (e.g.,
no random assignment, no alternative treatment given to
the control group, no assessment of the instructor's
fidelity
to treatment), percentage of external validity criteria
met,
number of critical flaws challenging a study's external va-
lidity (e.g., insufficient information about the sample of
participants or about how disability was defined and as-
sessed). Troia evaluated 28 of the studies included in our
database. We applied his ratings and rankings to the 56
comparisons derived from these studies. We did this
without checking Troia's evaluations for accuracy, so any
incorrect codings of the studies are the result of Troia's
procedures, not ours.
Four people coded the studies and entered values
into the SPSS
(Statistical
Package for the Social Sciences)
database. The reliability
of moderator-variable codes was
checked by comparing codes in the database to codes
generated by one of the coders who recoded 14 of the ar-
ticles (15%
of the cases). The percentage of agreement of
the codes was 94%.
All of the means, standard
deviations,
and sample sizes entered into the database were verified
at least twice for accuracy.
Meta-analysis. The primary
statistic used in our
analysis of outcomes was effect size, indicating whether
and by how much performance of the treatment group
exceeded performance of the control group, with the
difference expressed in standard deviation units. The for-
mula used to calculate raw effect sizes for each treatment-
control comparison was the mean of the treatment group
minus the mean of the control group divided by a pooled
standard deviation. Use of the pooled SD was adopted by
the National Reading Panel for use in all of its meta-
analyses.
Most of our studies reported treatment and control
group means and standard deviations, which were used
to calculate effect sizes. However, there were 14 studies
that lacked sufficient information. We employed DSTAT
(Johnson, 1989) to estimate these effects, usually from
F- or t- or MSE
values, or we obtained the information
from authors.
Some studies included multiple measures of PA,
reading, or spelling. To ensure that each treatment-control
comparison contributed only one effect size to each of
the three outcomes, we calculated a raw effect size (g)
for each measure when several were taken, and then we
averaged the effect sizes across these measures to create
one composite effect size for each outcome.
The analysis of effect sizes across comparisons was
conducted by giving more weight to effect sizes that
were based on larger samples of participants.
However, a
few studies administered instruction
to groups of students
and hence used groups rather than individual students as
the unit of analysis in their statistics (Byrne & Fielding-
Barnsley, 1991; Castle, Riach, & Nicholson, 1994;
O'Connor,
Jenkins, & Slocum, 1995; Torgesen, Morgan, &
Davis, 1992; Williams, 1980, Experiment 2). Using the
number of groups in the weighting procedure for these
comparisons had the effect of underrepresenting their ef-
fect sizes relative to effect sizes of the other comparisons
where the number of participants
was used. To address
this problem, we used the unit of analysis to convert raw
effect sizes (g) to corrected effect sizes (d) for each
treatment-control
comparison. Then when we calculated
mean effect sizes across comparisons, we weighted the
individual effect sizes (d) by the number of students in
the sample, not by the unit of analysis, thus ensuring that
no cases were underrepresented.
We employed the DSTAT statistical
package
(Johnson, 1989) to determine effect sizes and to test the
influence of moderator variables on effect sizes. Each
moderator variable had at least two levels. We tested
whether the mean weighted effect size (d) at each level
was statistically
greater than zero at p < .05, whether the
individual effect sizes at each level were statistically
ho-
mogeneous (p < .05), and whether effect sizes differed
statistically
at different levels of the moderator variables
(p < .05). We received assistance from two consultants in
the conduct of our meta-analysis, Blair
Johnson and
Harris
Cooper, both of whom have authored papers or
books on the subject.
Results
Were effect sizes greater than zero?
The effect size statistic measures how much the
mean of the PA-instructed
group exceeded the mean of
the control group in standard deviation units. An effect
size of 1 indicates that the treatment group mean was
one standard deviation higher than the control group
mean, revealing a strong effect of instruction.
An effect
Meta-analysis of phonemic awareness instruction 259
size of 0 indicates that treatment and control group
means were identical, showing that instruction had no ef-
fect. To judge the strength of an effect size, values sug-
gested by Cohen (1988) are commonly used. An effect
size of d = 0.20 is considered small, an effect size of d =
0.50 moderate, and an effect size of d = 0.80 or above
large. Translated into more familiar
terms, d = 0.20 indi-
cates that the treatment has moved the average child
from the 50th to the 58th percentile; d = 0.50 indicates
that the treatment has moved the average child to the
69th percentile; d = 0.80 indicates that the treatment has
moved the average child to the 79th percentile.
Mean effect sizes (d) associated with levels of the
moderator variables are reported in Table 2 for phonemic
awareness outcomes, in Table 3 for reading outcomes,
and in Table 4 for spelling outcomes. Values of d that
were statistically greater than zero are marked with an as-
terisk. Inspection of values in Tables 2 and 3 reveals that
all of the effect sizes involving PA and reading outcomes
were statistically greater than zero (p < .05). This indi-
cates that instruction was uniformly effective in teaching
PA and in facilitating
transfer to reading across all levels
of the moderator variables that were considered.
The overall effect size of PA instruction on the ac-
quisition of PA was large, d = 0.86, based on 72 compar-
isons. This is somewhat less than the effect size detected
by B&I
(1999) who reported d = 1.04 based on 36 com-
parisons. The overall effect size on reading was moder-
ate, d = 0.53, based on 90 comparisons. This is slightly
greater than the effect size reported by B&I who found d
= 0.44, based on 34 cases. Although not identical, the val-
ues are comparable, indicating that our larger database
did not alter earlier overall findings.
Inspection of spelling outcomes in Table 4 reveals
that all but three effect sizes were statistically
greater than
zero. This indicates that, across most levels of the moder-
ator variables, PA instruction transferred and improved
spelling skills more than alternative
forms of instruction
or no instruction.
Some of the studies evaluated effects on an out-
come not expected to be affected by PA instruction, per-
formance in math. Math posttests were administered
immediately after instruction in 12 comparisons and after
some delay in three comparisons. Results in Table 3
show that the effect size was statistically nonsignificant
and close to zero (d = 0.03), indicating that the effects of
PA instruction
were limited to literacy outcomes. These
findings argue against the operation of any halo or
Hawthorne effect explaining the positive effect sizes.
In sum, these findings lead us to conclude with
much confidence that phonemic awareness instruction is
more effective than alternative
forms of instruction or no
instruction in helping children acquire phonemic aware-
ness and in facilitating
transfer
of PA skills to reading and
spelling.
Were effect sizes homogeneous?
A homogeneity analysis calculates how probable it
is that the variance exhibited among the effect sizes
would be observed if only sampling error
was making
them different (Cooper, 1998). If found homogeneous,
the mean effect size can be interpreted as representative
of that set. When effect sizes are not homogeneous, the
next step is to examine whether moderator variables cre-
ate homogeneity, indicating their power to explain the
variance.
Tables 2, 3, and 4 report results of the test of homo-
geneity (p < .05) and 95%
confidence intervals to reveal
how variable effect sizes were. At the tops of the tables, it
is apparent that on the immediate outcome measures of
PA, reading and spelling, effect sizes were not homoge-
neous. Effect sizes involving follow-up measures of PA
and spelling outcomes were homogeneous, but follow-up
reading effect sizes were not. Thus, there is reason to ex-
amine moderator variables that may explain effects on
immediate outcomes of all three variables and on follow-
up tests of reading.
Did moderator variables influence effect sizes?
Effects of several moderator variables were exam-
ined (see list in Table 1). Some cautions should be ap-
plied in interpreting
the effects of moderators. When
comparing the size of effects associated with levels of
moderators, one must remember that PA instruction was
effective for almost all the levels examined, so one should
not mistakenly equate smaller effects with statistically
non-
significant effects.
Also when effect sizes are larger for some levels of
a moderator variable than for others, it is not clear that
the moderator caused PA instruction to vary in its impact
rather
than a third hidden factor confounded with the
moderator. Likewise, if a moderator exerts no differential
effect, it may be that a third factor obscured the differ-
ence. This renders conclusions about the influence of
moderators tentative and suggestive rather than definitive.
Another caution is that all 96 treatment-control
com-
parisons in our database did not contribute to the calcula-
tion of every effect size. Rather
the comparisons changed
across moderator variables, either because studies did not
report the information, or they did not assess that out-
come at that test point. Thus, one cannot assume that ef-
fect sizes represent the whole database.
Outcome measures. From Table 2, it is apparent that
PA instruction was effective in improving children's
phonemic awareness. The effect size after instruction
was
large (d = 0.86), and this value did not differ statistically
260 READING RESEARCH
QUARTERLY July/August/September 2001 36/3
Table 2 Phonemic awareness outcomes: Mean effect sizes (d) as a function of moderator
variables
and statistical
tests
to determine whether
effect sizes were statistically
greater
than zero at
p < .05,
were homogeneous
at
p < .05,
and differed
from one another at
p < .05. Effect sizes were measured
immediately
after instruction unless
labeled as follow-up
Moderator variables Number of Mean Homogeneous 95% Contrastsa
and levels cases d confidence interval
Time of posttest
Immediate 72 .86* No .79 to .92 1us
Follow-up 14 .73* Yes .61 to .85
Outcome measures of PA
Segmentation 51 .87* No .79 to .94 S = D > B
Blending 33 .61* No .52 to .69 S > O
Deletion 25 .82* No .73 to .91 B = O
Other 37 .72* No .64 to .81 D = O
Characteristics
of participants
Reading level
At risk 15 .95* No .76 to 1.14 A = N > D
Disabled 15 .62* No .48 to .75
Normal progress 42 .93* No .85 to 1.01
Grade
Preschool 2 2.37* No 1.93 to 2.81 P > K > 1 = 2
Kindergarten 39 .95* No .87 to 1.04
First 15 .48* Yes .31 to .64
Second-sixth 16 .70* Yes .56 to .83
Socioeconomic status
Low 12 1.07* No .93 to 1.20 ius
Middle and high 17 1.02* No .87 to 1.18
Language of instruction
English 61 .99* No .90 to 1.07 E > O
Other 11 .65* Yes .55 to .76
Characteristics
of PA instruction
Skills taught
One skill 18 1.16* No .96 to 1.36 1 = 2 > 3
Two skills 24 1.03* No .92 to 1.14
Three or more skills 30 .70* No .61 to .78
Blend and segment only 18 .81* No .67 to .95 /us
Three or more skills 30 .70* No .61 to .78
Use of letters
Letters
manipulated 39 .89* No .80 to .98 /us
Letters not manipulated 33 .82* No .73 to .91
Delivery unit
Individual child 24 .60* Yes .47 to .72 S > I = C
Small groups 35 1.38* No 1.26 to 1.50
Classrooms 13 .67* No .57 to .76
Length of instruction
1 to 4.5 hours 15 .61* Yes .41 to .81 5 = 10 > 1 = 20
5 to 9.3 hours 24 1.37* No 1.23 to 1.51
10 to 18 hours 9 1.14* No .97 to 1.32
20 to 75 hours 22 .65* No .56 to .74
Characteristics
of instructors
Classroom teachers 19 .78* No .70 to .87 RO > CT
Researchers and others 53 .94* No .84 to 1.03
Computers 8 .66* Yes .52 to .85 O > C
Others 64 .89* No .82 to .96
(Conltintued)
Meta-analysis of phonemic awareness instruction 261
Table 2 Phonemic
awareness outcomes
(continued)
Moderator variables Number of Mean Homogeneous 95% Contrastsa
and levels cases d confidence interval
Characteristics of design
Random assignment 33 .87* No .77 to .97 ns
Matched 18 .92* No .75 to 1.09
Nonequivalent 21 .83* No .73 to .92
Fidelity checked 29 .66* No .56 to .75 N > F
Fidelity not checked 43 1.02* No .93 to 1.11
Treated controls 38 .89* No .79 to .99 ns
Untreated controls 34 .83* No .75 to .92
Size of sample
9 to 22 students 15 1.37* No 1.09 to 1.66 9 = 31 > 24 = 56
24 to 30 students 22 .70* No .53 to .87
31 to 53 students 13 1.10* No .90 to 1.30
56 to 383 students 22 .82* No .74 to .89
Characteristics of study
Year of publication
1976-1985 10 .73* Yes .53 to .94 1991 > others
1986-1990 16 .72* No .59 to .85 1976 = 1986 = 1996
1991-1995 31 1.18* No 1.07 to 1.30
1996-2000 15 .70* No .59 to .81
Note. d = mean effect size; Homogeneous reports results of the statistical
test of whether the set of effect sizes was homogeneous; PA = phonemic awareness; ns = results
were not statistically
significant.
"a
Contrasts
portray results of post hoc pairwise statistical tests comparing levels of moderator variables. Symbols are initial letters/numbers of levels.
Indicates that effect size was statistically greater than zero at p < .05.
from
the effect size at the follow-up
test (d = 0.73), indi-
cating
no significant long-term
decline.
Thus,
PA
instruc-
tion taught
phonemic
awareness
very
effectively,
and
students retained
their skill after
instruction
ended.
Comparison
of specific
PA skills
acquired
during
instruc-
tion indicated that effects were statistically larger
for seg-
mentation
and deletion
outcomes than for blending.
The strong gains
in PA
transferred to reading,
and
effects
persisted
through
the second follow-up
test. As ev-
ident in Table
3, reading
outcome effect sizes were mod-
erate,
and the effect
size at the end of instruction
(d =
0.53) did not differ
statistically
from that at the first follow-
up test (d = 0.45).
A statistically significant
effect size was
still
present
but statistically
smaller at the second follow-
up test (d = 0.23).
The type of test made a difference. The
effect
size was statistically larger
on experimenter
tests
than on standardized tests.
Some studies measured
reading performance
with
pseudowords
to assess children's
ability
to decode unfa-
miliar words.
From Table
3, it is apparent
that PA instruc-
tion benefited
decoding
skill
(d = 0.56). Effects were
statistically greater
than
zero,
were moderate,
and did not
differ
statistically
on experimenter-devised
tests and stan-
dardized tests.
The effect of PA
instruction
on reading comprehen-
sion was assessed
in 20 comparisons.
From Table
3, it is
apparent
that the effect size was statistically
greater
than
zero (d = 0.34), indicating
that PA instruction
exerted
a
small-to-moderate
impact
on readers'
ability
to compre-
hend text.
PA
instruction
also transferred to spelling
(see Table
4). The effect size following
instruction
(d = 0.59) was
moderate and statistically
greater
than the effect sizes at
the two delayed
posttests
(d = 0.37 and 0.20), both of
which were statistically greater
than zero and did not dif-
fer statistically.
The effect
size was statistically larger
on
experimenter
tests than on standardized tests of spelling.
Characteristics
of students.
Three
types
of readers
were distinguished:
at risk,
disabled
(RD),
and normally
progressing
readers.
Younger
students
at risk
for
develop-
ing reading problems
and older disabled readers
have
been found to exhibit excessive
difficulty
manipulating
phonemes
in words
(Bradley
& Bryant,
1983;
Juel, 1988;
Juel,
Griffith,
& Gough,
1986),
so we were especially
in-
terested in the impact
of PA instruction on these readers.
A comparison
of PA outcomes
across
the three
reader
groups
revealed that
although
effect sizes were moderate
to large
in all cases,
they were statistically larger
for at-risk
readers
(d = 0.95) and normally
progressing
readers
(d =
0.93) than
for
disabled readers
(d = 0.62).
Children
at risk
gained
as much
PA,
according
to statistical
tests,
as
normally developing
readers,
indicating
that at-risk
read-
262 READING RESEARCH
QUARTERLY July/August/September 2001 36/3
Table 3 Reading outcomes: Mean effect sizes (d) as a function of moderator
variables and statistical
tests to determine
whether effect sizes were statistically greater
than zero at p < .05, were homogeneous at p < .05, and differed
from one another at p < .05. Effect
sizes were measured immediately
after instruction
unless labeled as follow-up
Moderator
variables Number of Mean Homogeneous 95% Contrastsa
and levels cases d confidence interval
Characteristics
of outcome measures
Time of posttest
Immediate 90 .53* No .47 to .58 Im = 1 > 2
First
follow-up 35 .45* No .36 to .54
Second follow-up 8 .23* No .11 to .34
Type of word test
Experimenter 58 .61* No .54 to .69 E > S
Standardized 37 .32* No .23 to .42
Type of pseudoword test
Experimenter 47 .56* No .48 to .64 ns
Standardized 8 .49* Yes .29 to .69
Reading comprehension 20 .34* No .21 to .46
Math achievement 15 .03
ns No -.11 to .16
Characteristics
of participants
Reading level
Immediate posttest
At risk 27 .86* No .72 to 1.00 A > D = N
Disabled 17 .45* Yes .32 to .57
Normal progress 46 .47* No .39 to .54
Follow-up posttest
At risk 15 1.33* No 1.10 to 1.56 A > D = N
Disabled 8 .28* Yes .10 to .46
Normal progress 12 .30* Yes .19 to .42
Grade
Preschool 7 1.25* No 1.01 to 1.50 P > K = 1 = 2
Kindergarten 40 .48* No .40 to .56
First 25 .49* Yes .36 to .62
Second-sixth 18 .49* Yes .35 to .62
Socioeconomic status
Low 11 .45* No .33 to .58 MH > L
Middle and high 29 .84* No .72 to .96
Language of instruction
Immediate posttest
English 72 .63* No .55 to .70 E > O
Other 18 .36* No .27 to .46
Follow-up posttest
English 17 .42* Yes .28 to .56 ns
Other 18 .47* No .35 to .59
Characteristics
of PA instruction
Skills taught
Immediate posttest
One skill 32 .71* No .58 to .84 1 = 2 > 3
Two skills 29 .79* No .69 to .89
Three or more skills 29 .27* Yes .19 to .35
Follow-up posttest
One skill 11 .55* Yes .37 to .73 2 > 1 > 3
Two skills 9 1.28* No .56 to .89
Three or more skills 15 .23* Yes .11 to .37
Blend and segment only 19 .67* No .54 to .81 BS > 3
Three or more skills 29 .27* Yes .19 to .35
Use of letters
Immediate posttest
Letters
manipulated 48 .67* No .59 to .75 L > NoL
Letters not manipulated 42 .38* No .30 to .46
Follow-up posttest
Letters
manipulated 16 .59* No .45 to .74 L > NoL
Letters
not manipulated 19 .36* No .25 to .47
(continued)
Meta-analysis of phonemic awareness instruction 263
Table 3 Reading
outcomes
(continued)
Moderator
variables Number of Mean Homogeneous 95% Contrasts
and levels cases d confidence interval
Delivery unit
Immediate posttest
Individual child 32 .45* Yes .34 to .57 S > I = C
Small groups 42 .81* No .71 to .92
Classrooms 16 .35* No .26 to .44
Follow-up posttest
Individual child 7 .33* Yes .11 to .55 S > I = C
Small groups 18 .83* No .66 to 1.00
Classrooms 10 .30* Yes .18 to .42
Length of instruction
1 to 4.5 hours 17 .61" Yes .42 to .79 1 = 5 = 10
5 to 9.3 hours 23 .76* No .62 to .89 20 < others
10 to 18 hours 19 .86* No .72 to 1.00
20 to 75 hours 25 .31* No .22 to .39
Characteristics
of instructors
Immediate posttest
Classroom teachers 22 .41* No .33 to .49 RO > CT
Researchers and others 68 .64* No .56 to .73
Follow-up posttest
Classroom teachers 12 .32* Yes .20 to .43 RO > CT
Researchers and others 23 .63* No .49 to .77
Computers 8 .33* Yes .16 to .49 O > C
Others 82 .55* No .49 to .61
Characteristics
of design
Random assignment 46 .63* No .54 to .72 R > N
Matched 22 .57* Yes .43 to .72 M = all
Nonequivalent 20 .40* No .31 to .49
Fidelity checked 31 .43* No .34 to .53 N > F
Fidelity not checked 59 .59* No .51 to .66
Control
group
Immediate posttest
Treated controls 54 .65* No .56 to .73 T > U
Untreated controls 36 .41* No .33 to .49
Follow-up posttest
Treated controls 20 .62* No .48 to .75 T > U
Untreated controls 15 .32* Yes .20 to .44
Size of sample
9 to 22 students 24 .72* No .51 to .92 9 = 31 > 56
24 to 30 students 22 .54* Yes .37 to .70 24 = 9,56
31 to 53 students 22 .91" No .76 to 1.05 31 > 24
56 to 383 students 22 .40* No .33 to .48
Characteristics
of Study
Year of publication
1976-1985 20 .77" No .62 to .93 1976 = 1991 >
1986-1990 16 .36* Yes .24 to .49 1986 = 1996
1991-1995 41 .77* No .67 to .87
1996-2000 13 .21* Yes .11 to .32
Note.
d = mean
effect
size;
Homogeneous reports
results
of the statistical test
of whether
the set of effect sizes was homogeneous;
PA
= phonemic
awareness;
Its
= results
were
not statistically significant.
"a
Contrasts
portray
results
of post
hoc pairwise
statistical
tests
comparing
levels
of moderator
variables.
Symbols
are initial
letters/numbers
of levels.
Indicates
that
effect size was statistically
greater
than zero
at
p < .05.
ers were not any less responsive to instruction
in acquir-
ing PA. Disabled readers may have exhibited smaller ef-
fect sizes because they were older and relatively more
advanced in PA skills with less room for gains than the
younger readers, and also because they were taught more
advanced forms of PA that may be harder
to acquire.
Transfer
of PA instruction to reading was also influ-
enced by reader ability. Table 3 reveals that at-risk chil-
dren showed statistically larger transfer effects on reading
(d = 0.86) than normal and RD students whose effect
sizes did not differ statistically
(d = 0.47 for normals and
d = 0.45 for RD). Effect sizes on follow-up reading tests
264 READING RESEARCH QUARTERLY July/August/September 2001 36/3
Table 4 Spelling outcomes: mean effect sizes (d) as a function of moderator variables and statistical
tests to determine
whether effect sizes were statistically greater
than zero at p < .05, were homogeneous at p < .05, and differed
from one another at p < .05. Effect
sizes were measured immediately
after
instruction
unless labeled as follow-up
Moderator
variables Number of Mean Homogeneous 95% Contrastsa
and levels cases d confidence interval
Characteristics
of outcome measures
Time of posttest
Immediate 39 .59* No .49 to .68 Im > 1 = 2
First
follow-up 17 .37* Yes .26 to .48
Second follow-up 6 .20* No .08 to .32
Type of spelling test
Experimenter 24 .75* No .62 to .89 E > S
Standardized 20 .41* No .29 to .53
Characteristics
of participants
Reading level
At risk 13 .76* No .54 to .98 A = N > D
Disabled 11 .15
ns Yes -.00 to .31
Normal progress 15 .88* No .74 to 1.02
Grade
Preschool 0 -
Kindergarten 15 .97* No .82 to 1.13 K > 1 > 2
First 16 .52* No .37 to .68
Second-sixth 8 .14
ns Yes -.04 to .33
Socioeconomic status
Low 6 .76* Yes .57 to .95 MH > L
Middle and high 9 1.17* No .88 to 1.47
Language of instruction
English 32 .60* No .49 to .70 1s
Other 7 .55* Yes .31 to .78
Characteristics of PA instruction
PA skills taught
One skill 17 .74* No .56 to .92 1 = 2 > 3
Two skills 12 .87* Yes .71 to 1.03
Three or more skills 10 .23* No .07 to .38
Blend and segment only 7 .79* Yes .49 to 1.09 BS > 3
Three or more skills 10 .23* No .07 to .38
Use of letters
Letters
manipulated 27 .61* No .50 to .72 L > NoL
Letters not manipulated 12 .34* No .25 to .42
Delivery unit
Individual child 14 .36* No .20 to .52 S > I
Small groups 20 .77* No .63 to .90 C = all
Classrooms 5 .56* No .33 to .78
Length of instruction
1 to 4.5 hours 0
5 to 9.3 hours 8 1.13* Yes .86 to 1.39 5 = 10 > 20
10 to 18 hours 10 .87* No .69 to 1.05
20 to 75 hours 18 .32* No .19 to .45
Characteristics
of instructors
Classroom teachers 9 .74* No .58 to .90 CT > RO
Researchers and others 30 .51* No .39 to .62
Computers 6 .09zns Yes -.10 to .28 O > C
Others 33 .74* No .63 to .85
(continued)
Meta-analysis of phonemic awareness instruction 265
Table 4 Spelling
outcomes
(continued)
Moderator variables Number of Mean Homogeneous 95% Contrastsa
and levels cases d confidence interval
Characteristics
of design
Random assignment 17 .37* No .23 to .50 M = N > R
Matched 12 .73* No .52 to .93
Nonequivalent 10 .86* Yes .69 to 1.04
Fidelity checked 15 .44* No .30 to .59 N > F
Fidelity not checked 24 .69* No .57 to .81
Treated controls 24 .43* No .30 to .55 U > T
Untreated controls 15 .82* No .67 to .96
Size of sample
9 to 22 students 15 .85* Yes .59 to 1.10 24 > all
24 to 30 students 3 1.68* Yes 1.15 to 2.21 9 > 56
31 to 53 students 8 .75* No .51 to .98 31 = 9,56
56 to 383 students 13 .45* No .34 to .56
Note. d = mean effect size; Homogeneous reports results of the statistical
test of whether the set of effect sizes was homogeneous; PA = phonemic awareness; ns = results
were not statistically
significant.
"a
Contrasts
portray results of post hoc pairwise statistical tests comparing levels of moderator variables. Symbols are initial letters/numbers of levels.
Indicates that effect size was statistically greater than zero at p < .05.
showed the same pattern except that the effect size for
at-risk
students grew even larger (d = 1.33) while the ef-
fect sizes for the other two groups grew smaller (d = 0.30
for normals and 0.28 for RD). These findings indicate that
PA instruction gave at-risk students a bigger boost in
reading than it gave normals or disabled readers.
Transfer
of PA instruction to spelling was also influ-
enced by reader ability. Effect sizes were large and did
not differ statistically
for at-risk
(d = 0.76) and normal
readers (d = 0.88), indicating that PA instruction strongly
benefited spelling for these students. However, the effect
size was small and not statistically
different from zero for
disabled readers (d = 0.15). Moreover, the set of effect
sizes for disabled readers was statistically
homogeneous,
indicating that no further
analysis of moderator variables
was needed to conclude that PA instruction did not help
RD students to spell.
The effects of PA instruction
were examined at vari-
ous grade levels: preschool, kindergarten,
first
grade, and
second-sixth grades. It should be noted that 78%
of the
second-sixth grade comparisons (i.e., 14 out of 18) in-
volved disabled readers, so findings apply mainly to RD
students and not to second-sixth graders in general. In
acquiring PA, preschoolers showed a very large effect
size (d = 2.37), although only two comparisons con-
tributed to this value, making it less reliable. The effect
on PA outcomes in kindergarten
(d = 0.95) was statistically
larger than the effect in first grade (d = 0.48) and in sec-
ond-sixth grades (d = 0.70). The latter two effect sizes
did not differ statistically.
These findings indicate that
preschoolers and kindergartners
gained the most PA, not
surprisingly
because they started out with the least PA.
Effect sizes for reading outcomes in Table 3 reveal
that PA instruction transferred
to reading to a similar ex-
tent for kindergartners,
first graders, and second-sixth
graders. Effect sizes were all statistically
greater than zero
and did not differ statistically
(ds = 0.48 to 0.49). The ef-
fect size for preschoolers was much larger statistically
(d
= 1.25), based on 7 comparisons and tested with simpli-
fied word recognition tests.
Transfer of PA instruction to spelling was statistically
greater among kindergartners
(d = 0.97) than among first
graders (d = 0.52). There was no transfer to spelling
among the second-sixth graders for whom the effect size
(d = 0.14) did not differ from zero statistically.
Spelling
was not measured in the preschool studies. The absence
of an effect on spelling among the older children arose
because all of the comparisons involved disabled readers
who did not gain in spelling from PA instruction.
SES levels did not have an impact on the acquisition
of PA, but they did influence reading and spelling out-
comes. As evident in Table 2, both low-SES and mid-high-
SES
groups showed large effect sizes that did not differ
significantly
in learning PA. However, transfer to reading
and spelling was statistically greater among mid-high than
among low-SES
students (see Tables 3 and 4). It might be
noted that most studies of disabled readers did not report
the students' SES,
so effect sizes are based primarily
on
the SES
of normally developing and at-risk children.
Studies examining PA instruction
were conducted
not only in English-speaking countries but also in coun-
tries speaking languages other than English. A compari-
son of effect sizes revealed that PA instruction exerted a
statistically
larger impact on the acquisition of PA by
266 READING
RESEARCH
QUARTERLY July/August/September 2001 36/3
English-speaking students (d = 0.99) than by the non-
English-speaking students (d = 0.65). Transfer
to reading
was also statistically greater for English-speaking students
(d = 0.63) than for others (d = 0.36) on the immediate
test but not on the follow-up test. Effect sizes on spelling
outcomes did not differ statistically
in the two language
groups. One possible reason is that 94%
of the RD com-
parisons were in the English pool, possibly suppressing
the English spelling effect size. When the effect size was
recalculated with the RD comparisons removed, a differ-
ence emerged. As evident in Table 5, the English compar-
isons yielded a statistically
larger effect size on spelling (d
= 0.95) than the non-English comparisons (d = 0.51). One
possible reason for the larger effect sizes in English may
be that the English writing system is not as transparent
in
representing phonemes as it is in the majority
of the
other languages, so explicit PA instruction may make a
bigger contribution to clarifying phoneme units and their
linkage to graphemes in English.
Because results of the language moderator variable
changed when RD comparisons were removed from the
analysis, we examined whether results might change for
other moderators having uneven distributions
of disabled
readers across their levels. Inspection of distributions re-
vealed some potential cases. Disabled readers were older
(mostly in Grades 2 through 6), they tended to receive PA
Table 5 Mean effect
sizes (d) with reading
disabled
comparisons
removed from the database and statistical
tests to
determine whether effect sizes were statistically
greater
than zero at
p < .05, were homogeneous
at
p < .05,
and differed
from
one another at
p < .05.
Moderator
variables Number of Mean Homogeneous 95% Contrastsa
and levels cases d confidence interval
Spelling outcomes
Grade
Preschool 0
Kindergarten 15 .97* No .82 to 1.13 K > 1
First 13 .66* No .48 to .85
Second-sixth 0
Language of instruction
English 22 .95* No .82 to 1.09 E > O
Other 6 .51* Yes .28 to .75
PA skills taught
One skill 14 .77* No .58 to .96 ns
Two skills 11 .89* Yes .72 to 1.05
Three or more skills 3 .93* No .52 to 1.33
Blend and segment only 6 .85* Yes .54 to 1.16 ns
Three or more skills 3 .93* No .52 to 1.33
Letter use
Letters
manipulated 17 1.00* Yes .85 to 1.15 L > NoL
Letters not manipulated 11 .57* No .37 to .76
Delivery unit
Individual child 8 1.00* No .71 to 1.28 I = S > C
Small groups 15 .94* Yes .78 to 1.10
Classrooms 5 .56* No .33 to .78
Length of instruction
1 to 4.5 hours 0 - - -
5 to 9.3 hours 8 1.13* Yes .86 to 1.39 ns
10 to 18 hours 8 .91* No .73 to 1.10
20 to 75 hours 9 .75* Yes .50 to 1.01
Instructor
Classroom teachers 8 .74* No .58 to .91 ns
Researchers and others 20 .96* No .79 to 1.14
Phonemic awareness outcomes
Letter use
Letters
manipulated 25 1.11* No .99 <