Content uploaded by Anne E Cunningham
Author content
All content in this area was uploaded by Anne E Cunningham on Feb 08, 2019
Content may be subject to copyright.
JOURNAL OF EXPERlhfENTAL CHILD PSYCHOLOGY 38,
175-190 (1984)
Assessing Phonological Awareness in Kindergarten Children:
Issues of Task Comparability
KEITH
E.
STANOVICH
Oakland University
ANNE
E.
CUNNINGHAM
University of Michigan
AND
BARBARA
B.
CRAMER
Oakland University
Ten different phonological awareness tasks were administered to a group of
kindergarten children whose reading ability was assessed 1 year later. The ex-
traneous cognitive requirements inherent in the tasks varied widely. The children’s
performance on three tasks that involved a rhyming response was at ceiling, and
these tasks did not correlate with subsequent reading progress. The other seven
measures were all moderately related to later reading ability and, employed in
sets, were very strong predictors. The relative predictive accuracy of the phono-
logical tasks was equal to or better than more global measures of cognitive skills
such as an intelligence test and a reading readiness test. The phonological tasks
had a large amount of common variance. Factor analysis revealed only one factor
on which all the nonrhyming phonological tasks loaded highly. The results bolster
the construct validity of phonological awareness, indicate considerable comparability
and interchangeability among the tasks used to measure the construct, and are
encouraging as regards the possible use of such tasks in predictive test batteries.
Researchers interested in the cognitive determinants of early reading
acquisition have increasingly focused on phonological awareness as a
The authors thank Cecilia Wiar, Principal, and the teachers and students of Clarkston
Elementary School, Clarkston, MI, for their cooperation. Anita Davison and Maryanne
Dedtick deserve special thanks for their enthusiatic participation. The authors thank Dorothy
Feeman and Ruth Nathan for their comments of the manuscript. Requests for reprints
should be sent to Keith E. Stanovich, Department of Psychology, Oakland University,
Rochester, MI 48063.
175
0022~0965/84 $3.00
Copyright Q 1984 by Academic Press, Inc.
All rights of reproduction in any tom reserved.
176
STANOVICH, CUNNINGHAM, AND CRAMER
potentially important variable. There is now a substantial body of evidence
indicating that tasks that in some way tap phonological awareness are
moderate to strong predictors of the speed with which children acquire
reading fluency in the early grades (Bradley & Bryant, 1978; Calfee,
Lindamood, & Lindamood, 1973; Fox & Routh, 1976, 1980, 1983; Golinkoff,
1978; Helfgott, 1976; Jorm & Share, 1983; Liberman, 1973, 1982; Rozin
& Gleitman, 1977; Treiman & Baron, 1981; Williams, 1980). The interest
in this particular cognitive skill has been fueled by recent evidence indicating
that the early development of phonological awareness is causally linked
to rapid reading acquisition (Bradley & Bryant, 1983; Perfetti, Beck, &
Hughes, 1981; Treiman & Baron, 1983; Williams, 1980).
A large number of different experimental paradigms have been used
to assess phonological awareness, including rhyming tasks, phoneme
segmentation tasks, matching tasks, phoneme substitution tasks, blending
tasks, and phoneme counting tasks, to name just a few (see Lewkowicz,
1980, for a useful typology). The plethora of tasks, however, has made
a consolidation of the knowledge gained from studies in this area very
difficult. All of the tasks that have been used involve many cognitive
processes (e.g., short-term memory, stimulus comparison, processing of
task instructions) in addition to the phonological analysis ability that is
the focus of interest. Without careful task analysis and comparison it
will remain unclear to what extent the predictive power of these tasks
resides in the phonological ability or the other extraneous cognitive
processes.
The present authors are not the first to recognize that a lack of direct
task comparisons is the current Achilles’ heel of the phonological awareness
literature. After a thorough review of the existing research Lewkowicz
(1980) observed that
There has been surprisingly little comparison, at least in print, of one phonemic
awareness task with another. There has been little analysis of similarities and
differences between tasks, of relative difficulty of
tasks
or of which tasks are
most closely related to the reading process and are most likely to facilitate learning
to read. In my opinion, this lack of in-depth analysis of phonemic awareness tasks
and their relationship to reading has resulted in the obscuring of some important
differences between the tasks, and, as a consequence, in the failure of researchers
to focus on the most important tasks and questions that need to be asked about
them. (pp. 686-687)
Three years later the situation remained much the same, and Backman
(1983) concluded from her results that
Tasks which on the surface appear to be measuring the same phenomenon may
in fact require different degrees of linguistic awareness, or may differ in their
cognitive requirements. . . We must not talk about phoneme segmentation per
se in relation to reading, but segmentation within the context of a particular task.
PHONOLOGICAL AWARENESS
177
. . . Obviously, ease of understanding task requirements is intimately related to
the phenomenon of “linguistic insight” we are interested in. (pp. 476-477)
Both Lewkowicz (1980) and Backman (1983) emphasized that differing
cognitive requirements could lead to a divergence in the results obtained
from different tasks. However, it is equally true that similar extraneous
cognitive requirements could lead to a convergence of results from tasks
that actually tap different aspects of phonological awareness (or that tap
the same
aspect to differing
degrees). Actually, the literature on phono-
logical awareness has shown considerable convergence despite the pleth-
ora of paradigms that have been used and the absence of extensive direct
task comparisons. The fact that a wide variety of tasks has converged
on similar conclusions in this area of research is encouraging. However,
the general absence of direct comparisons between tasks in the published
literature places arguments for convergence on shaky ground. Suspicion
will always remain that the convergence is spurious, the result of the
other cognitive and linguistic requirements of the phonological tasks.
Thus, it is essential that some attempt be made to directly assess the
relationships between phonological tasks and determine their degree of
convergence. Otherwise, the impact of the strong results previously ob-
tained in this research area may be muted due to lingering doubts about
construct validity. Also, as Lewkowicz (1980) noted, multivariate studies
of phonological awareness tasks would necessarily address another im-
portant question, that of potential task differences in predictive accuracy.
The present study attempted to address these issues. Ten different
phonological tasks were administered to a group of kindergarten subjects.
The tasks were of several different types, covering many of the categories
(e.g., word-to-word matching, rhyme recognition, phoneme deletion,
phoneme substitution, and identification of missing phoneme) included
in the classification system developed by Lewkowicz (1980). Some
tasks
required abstraction of the initial phoneme, while others focused on the
final phoneme. Finally, the critical linguistic construction that was used
in the instructions to the child varied across
tasks (e.g., “same,” “dif-
ferent,” “ not same”). In short, the phonological task domain was widely
sampled in terms of task type, location of phonological contrast, and
task instructions. Whether the differing cognitive requirements of the
tasks are reflected in the patterns of the performance relationships should
be revealed by correlational analyses. Conversely, the same correlational
analyses will give a rough indication of the degree to which these tasks
tap a similar underlying construct of phonological awareness.
It
should
be possible to detect performance convergences that are simply due to
similarities in extraneous task requirements, because the latter varied
between tasks to differing degrees.
The important issue of the relative predictive power of different pho-
178
STANOVICH, CUNNINGHAM, AND CRAMER
nological awareness tasks was addressed by obtaining a standardized
measure of reading ability on the same subjects at the end of first grade.
Thus, correlations between the performance on the phonological measures
in kindergarten and reading achievement at the end of first grade could
be assessed.
As more research attention focuses on the theoretical importance of
measures of phonological awareness, practitioners will naturally question
whether the relationships between these tasks and reading ability have
any degree of practical utility. For example, the question of how these
measures relate to other more global prereading assessment devices such
as readiness tests and intelligence tests, and how all of these measures
compare in predictive power will be raised. These questions were also
addressed in the present study by administering a standardized reading
readiness test and a standardized general intelligence test to the sample
of kindergarten subjects.
METHODS
Subjects
Fifty-eight subjects (29 males and 29 females) were recruited from
three kindergarten classrooms in a predominantly middle-class elementary
school. Nine subjects (4 males and 5 females) failed to follow the instructions
on several tasks, leaving a total of 49 subjects for subsequent analysis.
For example, several of these subjects scored zero on the substitute
initial consonant and rhyme supply tasks because they gave random
responses or semantic associates rather than rhymes. These subjects
completed some of the other more difficult tasks, suggesting that their
failure on the rhyme supply was due to a failure to understand the
instructions. The status of these subjects as multivariate outliers was
confirmed by using the objective methods described in Tabachnick and
Fidel1 (1983). The mean age of the 49 subjects was 6 years and 2 months
(SD = 4.4 months) at the time of testing. The children were administered
a battery of 10 phonological tasks in May of the school year by the same
experimenter. The prereading sections of the Metropolitan Readiness
Tests (Level 2, Form P) were administered to all of the subjects. The
mean prereading skills composite score was 47.9 (SD = 13.8; mean
percentile rank = 47%). Forty-six of the forty-nine children were also
administered the Otis-Lennon School Ability Test (Primary 1, Form R).
The mean score on the Otis-Lennon was 37.8 (SD = lO.l>, while the
mean School Ability Index (IQ) was 103.8 (SD = 16.4). The two stan-
dardized measures were administered in late May and early June, folIowing
administration of the 10 phonological tasks. Thirty-one of the subjects
(16 males and 15 females) were available for testing the following year.
In May these children were administered the Reading Survey Test (Form
PHONOLOGICAL AWARENESS
179
JS, Primary Level 1) of the Metropolitan Achievement Tests. The mean
raw score on this test was 42.3 (SD = 12.4) and the mean grade equivalent
was 3.1 (SD = 1.6).
Tasks and Procedure
Ten phonological awareness tasks were individually administered to
the subjects. Each task took approximately 10 min. to complete and was
administered on a separate day. The order of presentation was rhyme
supply, rhyme choice, initial consonant same, final consonant same, strip
initial consonant, substitute initial consonant, initial consonant not same,
final consonant different, initial consonant different, and supply initial
consonant. The subject’s score on each task was the total number of
correct responses, and the maximum score on each was 10. In tasks
involving a multiple choice the position of the correct alternative was
randomly determined and occurred with approximately equal frequency
in all positions. The 10 experimental trials for each task were preceded
by 3-5 practice trials during which the experimenter ensured that the
child understood the task. On these trials the experimenter gave feedback
on the correctness of the subject’s response. In the case of an incorrect
response the subjects were told the correct answer and why it was
correct. Following the practice trials the subjects were told that the
experimenter could no longer help them and that they were simply to
try their best. Subjects were also always told that if they did not know
the answer they were to guess.
Rhyme supply.
This task assessed children’s ability to provide a word
that rhymed with the target word. The 10 experimental words were
nose,
pup, sky, toy, hill, wing, mouse, tip, note,
and
look.
The words were
orally presented to the subjects. The subjects were told the experimenter
would say a word aloud to them and that their task was to provide
another word that rhymed or had the same ending sound as the target
word. For practice, the experimenter instructed the subject to listen to
the
wordsJish-dish.
The subjects were then told to say these words out
loud. The subjects were told that both words ended with the -ish sound
(that is, that they rhymed). Following the practice trials the experimenter
pronounced the 10 experimental words aloud to the subjects, and the
subjects responded orally.
Rhyme choice.
In this task subjects were provided with the stimulus
word and asked to choose 1 of 3 words that rhymed with it. The 10
experimental words were
star, mop, green, plane, clown, flash, cake,
jump,
box, andjeep. The subjects were told to listen closely to the target
word and the following 3 words. Their task was to choose a word that
rhymed with the target word. The experimenter began with two explicit
examples: “Listen to the word
pet. Now
saw the word
pet.
Tell me
which of these three words rhymes with
pet: barn, net, hand.” The
180
STANOVICH, CUNNINGHAM, AND CRAMER
experimenter then explained to the child how he was correct or incorrect
and the reasons why. Following the remaining practice trials, the
10
experimental trials were administered.
Initial consonant same.
This task consisted of 3 practice and 10 ex-
perimental multiple choice trials. The 10 experimental words were
milk,
pear, fan, bone, soap, tent, leg, duck, nest,
and
key.
In the practice
trials, subjects were instructed to listen closely to the beginning sound
of the target word. The subjects were then instructed to say the target
word aloud. The experimenter then pronounced the target word followed
by 3 words. Their task was to decide which had the same initial sound
as the target word.
Final consonant same.
This task consisted of 3 practice and 10 ex-
perimental multiple choice trials. Each trial had a target word followed
by 3 alternatives. The 10 experimental target words were
worm, cup,
pan, beat, 1eaJ bud, house, hook, nail,
and
bug.
The subjects were
instructed to listen closely to the target word and the 3 following words.
Their task was to choose the word that shared the same ending sound
as the target. A picture of the target word was provided for the subject.
The picture remained in front of the subject during the trial. It was
hypothesized that providing a picture of the target word would reduce
memory load for the subject. This procedure, however, was followed
only for this task. The experimenter provided the following example to
the subject, “Say the word
meat
aloud and listen to the ending sound.
If I say the word
meat
and thenfin,
coat, glass,
which word would you
say has the same ending sound as
meat?”
A picture of the target word
was placed on the table in front of the subject when the experimenter
said the word
“meat.”
The experimenter then explained to the subject
why the response was correct or incorrect and the remaining practice
trials and the 10 experimental trials were completed.
Strip initial consonant.
In this task subjects were required to delete
the initial phoneme of a word and pronounce the embedded word that
remained. This task was originally used by Bruce (1964) and Calfee,
Chapman, and Venezky (1972) and adapted for this experiment. The 10
experimental words were
pink, told, man, nice, win, bus, pitch, car, hit,
and
pout.
The experimenter instructed the subjects to listen closely to
the target word and then remove the first sound. The experimenter provided
the example, “Listen to the word
task.
If you take away the /t/ sound,
what word is left?” The subject was then told if he was correct or
incorrect and why. After the remaining practice trials the 10 experimental
trials were administered.
Substitute initial consonant.
This task required subjects to isolate the
initial sound of a word and then substitute a different sound to produce
a new word. The 10 target words were
top, bell, lip, fed, gum, sick, pin,
cat, sap,
and
cut.
The subjects were told that the experimenter was going
PHONOLOGICAL AWARENESS
181
to say a word, and that their task was to change the word by substituting
the initial sound thereby creating a different word. An example was
provided by the experimenter, “If I say the word go, and then change
the first sound by changing it to /n/ the new word will be IZO.” The
experimenter then asked the subject to try to change the initial phoneme
in the word hang. The subjects were told if they were correct or not
and the reasons why. Following this example and the remaining practice
trials the 10 experimental trials were administered.
Znitiul consonant difirent.
The subjects were told that the experimenter
would read four words out loud to them. Their task was to listen closely
to the beginning sound of each word and choose the word that had
beginning sound that was different from the other words. The experimenter
instructed, “Say the words
bug, nine, beach,
and
bike.
Can you tell me
which of these words has a different beginning sound-bag,
nine, beach,
bike?”
Following additional practice trials, the subjects were administered
the 10 experimental trials. The 10 correct words were
ear, pop, hill,
band, arm, give, van, cart, rice,
and
teeth.
Initial consonant not same.
This task is structurally similar to the
initial consonant different task except that the directions are phrased in
a negative manner. The 10 experimental target words were
boy, doll,
sun, kite, man, nest, fish, train, pie,
and
lamp.
The subjects were instructed
to listen closely to the initial sound of the first word. The experimenter
said, “I am going to say a word aloud followed by three more words.
Your task is to tell me which word does not begin with the same sound
as the first word. Say the word
mud.
Now say the words
mice, dig,
and
mouth.
Can you tell me which word did not have the same beginning
sound as
mud?”
Each subject was told if their answer was correct or
incorrect and the reasons why. Following the remaining practice trials
the 10 experimental trials were administered.
Final consonant different.
In this multiple choice task, subjects were
asked to identify one of four words which had a final sound that was
different from the others. Three practice and ten experimental trials were
administered. The three practice words were
rut, can,
and
log.
The 10
experimental target words were
hum, cup, leaf, flag, dress, wrist, ball,
sand, ruin,
and
desk.
The subjects were advised to listen closely to the
four words the experimenter would read to them. Their task was to
choose the word that had a different ending sound. For example, the
experimenter said “Listen to the following four words:
rut, dir,ze, bout,
mitt.
Say these words out loud. One of them has a different ending
sound. Can you tell me which word has a different sound at the end of
the word?” The experimenter then told the subjects when they were
correct or incorrect and the reasons why. Following the three practice
trials the subjects were informed that the experimenter could no longer
182
STANOVICH, CUNNINGHAM, AND CRAMER
help them and that if they could not choose the correct word that they
should give their best guess.
Supply initial consonant.
This task assessed the child’s ability to isolate
and produce the initial phoneme of a word. Subjects were orally presented
a pair of words that were identical except that the initial phoneme had
been deleted from the second word. The 10 experimental word pairs
were
meal-eel, Jill-ill, sit-it, land-and, near-ear, pair-air, bend-end,
task-ask, date-ate,
and
can’t-ant.
The subjects were told that they
would be hearing two words that were the same except for the beginning
sound. For example, the experimeter said “Say the word
cut. Now
say
at.
What sound do you hear in
cut,
that is missing from
at?”
The subjects
were told the correct answer if they were unable to respond correctly.
Following the remaining practice trials the 10 experimental trials were
administered.
RESULTS
Descriptive statistics on each of the 10 phonological awareness measures
are presented in Table 1. Hotelling’s
T,
calculated with the 10 phonological
tasks as the set of dependent measures, indicated no sex differences in
the data
(T
= 8.614,
F
< l), so this variable will not be considered
further. A perusal of the means reveals several patterns. The strip initial
consonant task was the most difficult task. The mean score on this task
was extremely low and the scores were positively skewed. The mode
was zero (a score attained by 25 children), but 10 children attained a
score of 8 or above. Thus, there is some indication of a bimodal distribution
of responses, a pattern that has been observed before with the strip
consonant task (see Calfee et al., 1972). Clearly this task exceeds the
cognitive and phonological analysis capabilities of many kindergarten
children. Nevertheless, the task was a moderately good predictor of first-
grade reading and did correlate with other tasks that had more symmetrical
distributions of responses (see below). The supply initial consonant task
and both measures involving the final consonant were relatively difficult
tasks (although note that the means are not directly comparable because
the multiple choice nature of the latter two tasks probably restricts guessing
responses).
Tasks where the critical sound contrast was at the beginning of the
word were easier than those where the critical sound was at the end of
the word. There was some statistical support for this trend. The mean
in the initial consonant same task was higher than the mean in the final
consonant same task,
t(48)
= 5.88,
p
< .OOl, and the mean in the initial
consonant different task was significantly higher than the mean in the
final consonant different task,
t(48)
= 3.57,
p
< .Ol. Marsh and Mineo
(1977) also found that phoneme isolation performance was superior when
the critical phonemic contrast was in the initial position, but this pattern
TABLE 1
PHONOLOGICAL AWARENESS MEASURES: DE~CIUFTIVE STATISTICS
Task Mean
Strip initial consonant 2.53 3.67 1.16 .95 .59 .70
Supply initial consonant 5.57 4.29 -.14 .95 .68 .79
Initial consonant same 7.29 2.46 - .57 .83 .62 .76
Initial consonant different 5.80 2.64 .I8 .86 .71 .84
Initial consonant not same 6.02 2.29 .08 .73 .74 .87
Fii consonant same 5.31 2.43 .46 .72 64 .71
Final consonant different 4.82 2.44 .57 .63 .68 .84
Rhyme choice 7.71 1.91 -.66 .65 .35 .54
Rhyme supply 8.57 1.26 - .71 30 .20 .36
Substitute initial consonant 8.63 1.33 - .77 .91 .13 .lO
Standard
deviation Skewness Reliability FL Metropolitan g
correlation 2
.42
.52 g
0
.39 ij
.60 F
.51
40 z
.45 %
.30
.ll #
.09 2
Note. SMC = squared multiple correlation of each variable with all other variables. FL = factor loading.
184
STANOVICH, CUNNINGHAM, AND CRAMER
interacted with phoneme class (continuants versus stops). A post hoc
analysis of our data revealed no such interaction, error rates being very
similar for continuants and stops in all tasks.
The three easiest tasks were substitute initial consonant, rhyme supply,
and rhyme choice. Note that the substitute initial consonant task essentially
requires the subject to produce a rhyme. It differs from the other two
tasks only in the phrasing of the instructions, which do not explicitly
mention rhyming. The superior performance on the three rhyming tasks
confirms previous research, and the reports of teachers that children
often enter school with rhyming skills and that it is the easiest auditory
analysis task to teach (see Jusczyk, 1977; Lewkowicz, 1980). Indeed,
there is strong evidence that performance on these three tasks was at
ceiling. The rhyming tasks had the three smallest standard deviations
and the distributions of all three were characterized by negative skewness.
The distribution of responses on the substitute initial consonant task
illustrates the ceiling effect. The modal score was 10 (achieved by 16 of
the subjects), and the next most frequent score was 9 (achieved by 14
subjects). The ceiling effects apparent in the three rhyming tasks probably
account for the fact that these tasks were poor predictors of later reading
achievement and were only weakly related to the other phonological
awareness measures (see below). Essentially, they suffer from extremely
restricted ranges, as indicated by the standard deviations in Table 1, and
thus will necessarily fail to correlate with other variables.
The split-half reliability (Spearman-Brown corrected) of each task is
also presented in Table 1. In general, the reliabilities of the 10 tasks
ranged from low moderate to high. The mean task reliability was .81.
Of course, the magnitudes of the correlations displayed by a given task
are limited by the reliability of the task, and this fact should be remembered
when interpreting the obtained correlations. In general, however, a con-
sideration of the reliabilities does not attenuate the conclusions drawn
here, and in many cases consideration of the reliabilities serves to strengthen
the apparent trends. For example, some tasks that were good predictors
and that were strongly related to other tasks were only moderately reliable
(initial consonant not same), while others that displayed weak relationships
were highly reliable (rhyme supply and substitute initial consonant, although
the high reliability may also be due to the ceiling effects in the performance
on these tasks).
The interrelationships among the phonological awareness tasks are
displayed in Table 2 which is a correlation matrix of the ten measures.
All correlations larger than .28 are significant at the .0.5 level. The major
data pattern is apparent from a visual inspection of the matrix. The seven
nonrhyming tasks show moderate to strong relationships with each other.
The substitute initial consonant task does not correlate with anything.
The rhyme choice and rhyme supply tasks show weak correlations with
PHONOLOGICAL AWARENESS
185
TABLE 2
INTERCORRELATIONS OF ALL TASKS
Task 2 3 4 5 6 7 8 9 10
1. Strip initial consonant
2. Supply initial consonant
3. Initial consonant same
4. Initial consonant different
5. Initial consonant not same
6. Final consonant same
7. Final consonant different
8. Rhyme choice
9. Rhyme supply
10. Substitute initial consonant
.49 .45 .57 .62 .66 .66 .30 .18
.62 .68 .73 .41 .69 .47 .35
.62 .74 .53 .60 .48 .20
.70 .68 .72 .40 .28
55 .71 .44 .34
.61 .39 .24
.40 .26
.32
- .02
.Ol
.09
.17
.11
.06
- .02
.16
.13
the other variables. This pattern is quantified in Table 1, which displays
the squared multiple correlations of each variable as a dependent variable
regressed on the other nine. The values for the seven nonrhyming tasks
were very high, approaching their reliabilities in some cases. The seven
nonrhyming tasks appear to have much common variance. In contrast,
the SMCs of the rhyming tasks were very low (considering the number
of predictor variables), ranging from a high of .35 for rhyme choice to
a low of .13 for substitute initial consonant. Basically, the extremely
restricted range of the scores on the rhyming tasks prevents them from
correlating with anything.
The uniformly moderate to high correlations among the seven nonrhyming
tasks was somewhat surprising given their different cognitive requirements
and moderate reliabilities in some cases. The mean correlation between
these seven variables was .62, which is quite high considering that their
mean reliability is .8 1. Also surprising was the relative lack of clusters
among the correlations of the seven nonrhyming tasks. The variables
were relatively uniformly intercorrelated. These visual impressions from
Table 2 were confirmed by a principle factor analysis carried out on the
data of all ten tasks. With squared multiple correlations serving as initial
commonality estimates, only the first factor exceeded the eigenvalue>l
criterion (the eigenvalue of the second factor was .398). This factor
accounted for 47.8% of the total variance in the variables. A stable
solution was reached after four iterations, and the factor loadings from
this solution are displayed in Table 1. Predictably from the pattern of
correlations in Table 2, the seven nonrhyming tasks had high loadings
on the first principle factor, the rhyme choice task had a moderate loading,
and the other two rhyming tasks had low loadings.
The last column of Table 1 displays the correlation between each of
the phonological measures and the score on the Reading Survey Test of
the Metropolitan Achievement Tests that was administered at the end
186
STANOVICH, CUNNINGHAM, AND CRAMER
of first grade. Correlations larger than .35 are significant at the .05 level.
All seven of the nonrhyming tasks displayed significant correlations of
moderate strength. Some idea of the relative magnitude of the difference
in performance on the phonological tasks between readers of different
abilities is provided by Table 3. There the 31 subjects who were assessed
in the first grade have been split into a group of 16 less-skilled readers
and a group of 15 skilled readers based on their score on the Reading
Survey Test of the Metropolitan. The scores of the two groups were
significantly different at the A01 level. The mean grade equivalent of the
less-skilled group was 1.9 and the mean grade equivalent of the skilled
group was 4.3. From Table 3 it is apparent that all of the measures except
two of the rhyming tasks displayed significant differences between the
two groups. It is interesting that the skilled group is near ceiling even
on some of the nonrhyming tasks, whereas the performance of the less-
skilled group is far from ceiling on any of these tasks. The median split
thus gives some indication of the large performance disparity that is
present on some of the nonrhyming tasks.
In attempting to assess how to best characterize the ability of the
phonological tasks to predict first-grade reading levels, it is useful for
comparative purposes to consider the results involving the Metropolitan
Readiness Tests and Otis-Lennon IQ, two omnibus measures of cognitive
skills that were also administered to these children in kindergarten. The
prereading readiness test correlated 52 with first-grade reading ability
and the correlation between the raw score on the Otis-Lennon and
reading ability was .25. Thus, three of the seven nonrhyming phonological
measures (each containing 10 items and taking less than 10 min to ad-
minister) predicted first-grade reading ability as well as a standardized
readiness test containing six ditferent subsections, and all of the nonrhyming
phonological measures were superior to an omnibus IQ test. The mean
TABLE 3
MEAN SOURCES ON THE FHONOL~GICAL TASKS AS A FUNCTION OF READER SKILL
Task Skilled Less skilled
t
Strip initial consonant 4.13
Supply initial consonant 9.27
Initial consonant same 9.20
Initial consonant different 8.47
Initial consonant not same 8.07
Final consonant same 7.00
Final consonant different 6.93
Rhyme choice 9.07
Rhyme supply 9.00
Substitute initial consonant 8.87
* p <
.Ol.
1.13 2.80*
4.69 3.81*
7.00 3.43*
4.75 5.78*
5.56 4.52*
4.31 3.62*
4.06 4.44*
7.44 2.82*
8.75 0.57
8.38 0.92
PHONOLOGICAL AWARENESS
187
correlation between performance on the seven nonrhyming phonological
tasks and performance on the Metropolitan prereading readiness section
was .68. The analogous mean correlation with the Otis-Lennon score
was .55.
The trends already described were confirmed in a somewhat different
way by regression analyses. A stepwise regession analysis with first-
grade Metropolitan score as the dependent variable and the 10 phonological
variables as predictors confirmed the essential redundancy and large
variance overlap in the phonological measures. After initial consonant
different and initial consonant not same were entered into the equation
no other variable made a significant contribution to explaining reading
variance. Together, these two variables accounted for 66.2% of the variance
in reading ability (adjusted
R* =
.638). Both regression coefficients were
significant in the final analysis. The relatively high proportion of variance
explained was not primarily due to the optimization criteria of the stepwise
regression. The median proportion of variance explained by the 21 different
pairs of nonrhyming phonological tasks was 57.6%. This figure is com-
parable to the 59.1% of the variance in first-grade reading ability that is
accounted for when both the Otis-Lennon and Metropolitan readiness
test were employed as predictors. An additional hierarchical regression
analysis indicated that after the two phonological measures had been
entered into the regression equation the two standardized measures,
when entered as a set, did not account for a signScant additional proportion
of variance.
DISCUSSION
Descriptively, the ten phonological tasks broke down into three groups.
Three tasks that required a rhyming response were very easy. Ceiling
effects were apparent on these three tasks. The distribution of responses
was negatively skewed and the variance was highly restricted. As a result
of the restriction of range these three tasks displayed low correlations
with the other phonological tasks and with first-grade reading ability.
One task (strip initial consonant) was extremely difficult. The distribution
of responses on this variable was positively skewed and displayed a
tendency toward bimodality. The other six measures were of intermediate
difficulty and had distributions of responses that were more nearly
symmetrical.
The seven nonrhyming tasks were highly interrelated. Despite the
differing task requirements there was every indication that these tasks
were tapping a similar construct. The mean correlation between the seven
tasks was quite high, considering the reliability of the measures. The
squared multiple correlations for each of these variables was also quite
high. The visual impression that the correlations in Table 2 are uniformly
moderate and do not tend to cluster was confirmed by the factor analysis,
188
STANOVICH, CUNNINGHAM, AND CRAMER
which extracted only one factor upon which each of the seven variables
loaded highly. Regression analyses predicting first-grade reading ability
confirmed the essential redundancy of these seven variables. Stepwise
regression stopped after only two variables had entered the equation.
The proportion of variance accounted for did not change appreciably
when pairs of variables not chosen by the stepwise procedure were used
as predictors. Although caution in interpreting multivariate statistics is
always advised when the subjects to variables ratio is in the range employed
in our investigation, the convergence of all of the analyses suggests that
the data patterns are probably robust.
Performance on each of the seven nonrhyming tasks was related to
first-grade reading ability. The absolute magnitude of the performance
difference between skilled and less-skilled readers is apparent in Table
3 and in many cases is quite large, considering that the partitioning
represents not an extreme groups comparison but instead is a median
split. At the end of kindergarten the skilled readers were near ceiling on
several tasks, while the less-skilled readers were getting barely half of
the items correct.
The correlational data also suggest that the seven nonrhyming tasks
are quite impressive predictors of first-grade reading ability. All of the
correlations with Metropolitan scores were significant and at least of
moderate strength. The magnitude of the correlations must be considered
in the context of the reliability of the tasks (each administered in a very
short time span) and in the context of the type of correlations that are
obtained when more comprehensive and carefully standardized measures
of cognitive skills are employed. From this standpoint, the diagnosticity
of the phonological measures was truly impressive. All seven nonrhyming
measures correlated with first-grade reading more strongly than did a
standardized IQ test (see Stanovich, Cunningham, & Feeman, 1984,
where a similar result obtained when all the tests were administered at
the end of first grade). Three of the phonological tasks displayed correlations
with first grade reading as large as those obtained from a standardized
readiness test that was designed to tap a variety of reading-related cognitive
skills. A stepwise regression of the Metropolitan scores on the phonological
variables indicated that the latter explained 66.2% of the variance. When
each possible pair of nonrhyming phonological measures served as pre-
dictors the median proportion of variance explained was 57.6%. This
compares to 59.1% explained by the Metropolitan readiness test and
Otis-Lennon IQ in combination.
The results of this investigation bode well for the future use of pho-
nological awareness measures in both research and educational settings.
The wide variety of tasks that have been employed appear to be tapping
a similar ability. Results from different investigations are probably not
too contaminated by disparate task requirements. In fact, the degree of
PHONOLOGICAL AWARENESS
189
task convergence was quite a surprise to the present investigators who
entered upon this investigation fully prepared to uncover the fact that
the variance introduced by differing cognitive requirements would tend
to obscure the underlying phonological abilities that were being tapped
and would lead to vastly different patterns of task performance.
Finally, the uniformly moderate predictive accuracy of each task, coupled
with the impressive predictive power when sets of these measures are
used together, is an encouraging sign as regards future practical applications.
While we must reiterate the caution that the relationship between reading
ability and phonological awareness seems to be characterized by reciprocal
causation (see Baron & Treiman, 1980; Ehri, 1979; Morais, Cat-y, Alegria,
& Bertelson, 1979; Perfetti et al., 1981), the causal connection at the
earliest stages of reading acquisition is probably most strong from pho-
nological awareness to increased reading acquisition (Bradley & Bryant,
1983; Treiman & Baron, 1983).
REFERENCES
Backman, J. (1983). The role of psycholinguistic skills in reading acquisition: A look at
early readers.
Reading Research Quarterly, 18, 466-479.
Baron, J., & Treiman, R. (1980). Use of orthography in reading and learning to read. In
J. Kavanagh & R. Venezky (Eds.),
Orthography, reading, and dyslexia.
Baltimore:
Univ. Park Press.
Bradley, L., & Bryant, P. (1978). Difficulties in auditory organization as a possible cause
of reading backwardness.
Nature (London), 271, 746-747.
Bradley, L., & Bryant, P. (1983). Categorizing sounds and learning to read: A causal
connection.
Nature (London), 301, 419-421.
Bruce, D. (1964). The analysis of word sounds by young children,
British Journal of
Educational Psychology, 34,
158-170.
Calfee, R., Chapman, R. & Venezky, R. (1972). How a child needs to think to learn to
read. In L. Gregg (Ed.),
Cognition in learning and memory.
New York: Wiley.
Calfee, R., Lindamood, P., & Lindamood, C. (1973). Acoustic-phonetic skills and reading:
Kindergarten through twelfth grade.
Journal ofEducational Psychology, 64, 293-298.
Ehri, L. (1979). Linguistic insight: Threshold of reading acquisition. In T. Walker & G.
Mackinnon (Eds.),
Reading research: Advances in research and theory (Vol.1). New
York: Academic Press.
FOX, B., & Routh, D. (1976). Phonemic analysis and synthesis as word attack skills.
Journal of Educational Psychology, 68, 70-74.
FOX, B., & Routh, D. (1980). Phonemic analysis and severe reading disability. Journal of
Psycholinguistic Research, 9,
115-l 19.
FOX, B., & Routh, D. (1983). Reading disability, phonemic analysis, and dysphonic spelling:
A follow-up study.
Journal of Clinical Child Psychology, 12, 28-32.
Golinkoff, R. (1978). Phonemic awareness skills and reading achievement. In F. Murray
& J. Pikulski (Eds.),
The acquisition of reading.
Baltimore: Univ. Park Press.
Helfgott, J. (1976). Phonemic segmentation and blending skills of kindergarten children:
Implication for beginning reading acquisition.
Contemporary Educational
Psychology,
1,
157-169.
JOIYII, A., & Share, D. (1983). Phonological recoding and reading acquisition.
Applied
Psycholinguistics, 4,
103-147.
Jusczyk, P. (1977). Rhymes and reasons: Some aspects of the child’s appreciation of poetic
form.
Developmental Psychology, 13, 599-607.
190
STANOVICH, CUNNINGHAM, AND CRAMER
Lewkowicz, N. (1980). Phonemic awareness training: What to teach and how to teach it.
Journal of Educational Psychology, 72, 686-700.
Liberman, I. (1973). Segmentation of the spoken word and reading acquisition.
Bullerin
of the Orton Society, 23, 65-77.
Liberman, I. (1982). A language-oriented view of reading and its disabilities. In H. Myklebust
(Ed.),
Progress in learning disabilities (Vol.
5). New York: Grune 8t Stratton.
Marsh, G., & Mineo, R. (1977). Training preschool children to recognize phonemes in
words.
Journal of Educational Psychology, 69, 748-753.
Morais, J., Cary, L., Alegria, J., & Bertelson, P. (1979). Does awareness of speech as a
sequence of phones arise spontaneously?.
Cognition, 7,
323-331.
Perfetti, C., Beck, I., & Hughes, C. (1981, March).
Phonemic knowledge and learning to
read.
Paper presented at The meeting of the Society for Research in Child Development,
Boston.
Rozin, P., & Gleitman, L. (1977). The structure and acquisition of reading: II. The reading
process and the acquisition of the alphabetic principle. In A. Reber & D. Scarborough
(Eds.),
Toward a psychology of reading.
Hillsdale, NJ: Erlbaum.
Stanovich, K., Cunningham, A., & Feeman, D. (1984). Intelligence, cognitive skills, and
early reading progress.
Reading Research Quarterly,
19, 120-139.
Tabachnick, B., & Fidell, L.
(1983). Using multivariate statistics.
New York: Harper Jr
Row.
Treiman, R., & Baron, J. (1981). Segmental analysis ability: Development and relation to
reading ability. In T. Waller & G. Mackinnon (Ed%),
Reading research: Advances in
theory and practice (Vol. 3).
New York: Academic Press.
Treiman, R., & Baron, J. (1983). Phonemic-analysis training helps children benefit from
spelling-sound rules. Memory
& Cognition, 11, 382-389.
Williams, J. (1980). Teaching decoding with an emphasis on phoneme analysis and phoneme
blending.
Journal of Educational Psychology, 72,
1-15.
RECEIVED:
October 17, 1983;
REVISED:
January 11, 1984.
