Effects of prosodic modeling and repeated reading on poor readers' fluency and comprehension

Abstract

Repeated reading of meaningful text has been shown to produce improvements in reading rate, fluency, and comprehension in readers of varying ability. The assisted repeated reading (ARR) method, which provides a fluent and expressive (i.e., prosodic) model, has been proposed as being particularly helpful in this regard. However, it is unclear which component of the ARR method (prosodic modeling or reading practice with intact text) is the most influential factor. The present study examined the effects of text practice and prosodic modeling on the reading rate, accuracy, expressiveness, and comprehension of 40 grade 5 disabled readers. Text practice and prosodic modeling were systematically varied to create four training conditions. Each subject read the first half of a set of stories three times under one of the four experimental conditions. Pretest and posttest measures of the dependent variables were analyzed for both the training passages and the second half of each story, on which no training occurred (transfer passages). While reading performance improved across all conditions, substantial additional gains were produced by the conditions that included the practice of intact text. Modeling of prosody did not produce significant additional gains. Transfer effects were limited, with only the ARR condition producing improved accuracy on the second half of the stories.

Full text

Applied Psycholinguistics 17 (1996), 59-84
Printed in the United States of America
Effects of prosodic modeling and
repeated reading on poor readers'
fluency and comprehension
ARLENE R. YOUNG, PATRICIA GREIG BOWERS,
and G. E. MACKINNON
University of
Waterloo,
Ontario
ADDRESS FOR CORRESPONDENCE
Arlene Young, Developmental Neuropsychiatry Clinic, Child and Family Studies Centre,
Clark Institute of Psychiatry, 250 College St., Toronto, Ontario, Canada M5T 1R8
ABSTRACT
Repeated reading of meaningful text has been shown to produce improvements in reading rate,
fluency, and comprehension in readers of varying ability. The assisted repeated reading (ARR)
method, which provides a fluent and expressive (i.e., prosodic) model, has been proposed as
being particularly helpful in this regard. However, it is unclear which component of the ARR
method (prosodic modeling or reading practice with intact text) is the most influential factor.
The present study examined the effects of text practice and prosodic modeling on the reading
rate,
accuracy, expressiveness, and comprehension of 40 grade 5 disabled readers. Text prac-
tice and prosodic modeling were systematically varied to create four training conditions. Each
subject read the first half of a set of stories three times under one of the four experimental
conditions. Pretest and posttest measures of the dependent variables were analyzed for both
the training passages and the second half of each story, on which no training occurred (transfer
passages). While reading performance improved across all conditions, substantial additional
gains were produced by the conditions that included the practice of intact text. Modeling of
prosody did not produce significant additional gains. Transfer effects were limited, with only
the ARR condition producing improved accuracy on the second half of the stories.
Repeated reading of a meaningful text produces improvements in various
indices of reading skill, such as accuracy, fluency, and comprehension
(e.g., Bowers, 1993; Dowhower, 1987; Herman, 1985;
O'Shea,
Sindelar, &
O'Shea,
1985; Samuels, 1979). Various methods of rereading practice have
been described in the literature and applied to classroom practice (see
Downs
&
Morin, 1990). These include unassisted repeated reading, where a
child rereads a particular story or selection of stories to him- or herself
without adult assistance, and assisted repeated reading, where the child
reads in unison with a fluent oral model. Variations on the assisted repeated
reading method include reading-while-listening (McMahon, 1983), where
the child reads silently while listening to a tape-recorded reading, and the
neurological impress method (Heckelman, 1969), which is much like read-
© 1996 Cambridge University Press 0142-7164/96 $9.00 + .10

Applied Psycholinguistics 17:1
60
Young
et
al.: Effects
of
modeling and repeated reading
ing-while-listening except that the child and the model read in unison and
the model sits behind the child and uses finger pointing on the text to help
the child follow along.
While various methods of reading practice have been compared (e.g.,
Conte & Humphreys, 1989; Krug, Davis, & Glover, 1990; Reitsma, 1988),
few studies have reported findings that attempt to distinguish specifically
between the effects of spoken or modeled text and text repetition practice.
Notable exceptions include Rasinski (1990) and Dowhower (1987). Rasin-
ski reported significant but equivalent gains in rate and accuracy resulting
from repeated reading and listening-while-reading practice in a sample of
predominantly high-ability, grade 3 readers. Similarly, Dowhower (1987)
reported few differences from these methods in rate, accuracy, and compre-
hension among grade 2 "transitional readers" (i.e., those whose word recog-
nition skills were age appropriate, but who were slow readers). However,
differential improvements in the prosodic characteristics of oral reading
were noted by Dowhower in the group provided with a prosodic model.
Although the repeated reading method has been shown to facilitate chil-
dren's fluency, the relative contribution of text rereading practice versus
prosodic modeling remains unclear. While the identification of the effective
components of repeated reading practice has clear pedagogical applica-
tions,
such a determination also has implications for more theoretical ques-
tions regarding models of fluency development. In particular, models such
as LaBerge and Samuels's (1974) automaticity theory and Perfetti and Les-
gold's (1979) verbal efficiency theory (see also Lesgold & Perfetti, 1981)
contend that fluency and comprehension are limited in poor readers primar-
ily because of very weak skills at the single-word identification level. Ac-
cording to these models, slow and inefficient processing at the word level
results in few processing resources being available to integrate the meanings
of strings of words in understanding the text as a whole. Thus, gains shown
from repeated reading reflect improvements in the ease and speed of single-
word identification which, in turn, allow for a composite interpretation of
the text, which is reflected in better comprehension (and, one would as-
sume, improved expressiveness and fluency). While neither the verbal effi-
ciency nor automaticity models address the issue of prosodic modeling, the
emphasis on single-word identification skills within these theories suggests
that such modeling would be expected to be of limited benefit.
In contrast, Schreiber (1980, 1987) proposed that providing a model of
prdsodic reading may help to direct the reader's attention to cues of syntac-
tic structure that are contained in text - cues of which they are not yet fully
aware. Schreiber contended that children's reliance on prosodic cues to
process syntactic structure in speech accounts for an important difficulty
that beginning readers encounter in acquiring reading fluency. This position
can be summarized as follows: prosodic and rhythmic characteristics of
oral language are an important source of information regarding syntactic
segmentation. An implicit developmental assumption underlying this ac-
count is that the reliance on surface cues of language (such as prosody)
decreases as linguistic competence increases. Consequently, young children

Applied Psycholinguistics 17:1 . 61
Young et al.: Effects of modeling and repeated reading
and those with delayed linguistic development are expected to be particu-
larly reliant on prosodic cues for processing the syntactic structure of
speech. Schreiber suggested that this has particular implications for the
development of reading fluency because a text does not contain prosodic
cues to help children recognize phrasal boundaries.
According to Schreiber, learning to compensate for this essential differ-
ence between processing text and speech underlies the progression from
Samuels's (LaBerge & Samuels, 1974) second stage in reading fluency ac-
quisition (accurate, but slow word identification) to stage three (the auto-
maticity stage, where attention can be directed toward comprehension).
Hence, "stage-two readers must tacitly come to recognize the need to com-
pensate for the absence of graphic signals corresponding to the prosodic
cues and must make better use of the morphological and syntactic cues that
are preserved in order to move to automaticity" (Schreiber, 1987, p. 264).
The expectation from this point of view is that prosodic modeling and
practice with intact text will assist readers to bridge the gap between oral
and written language and allow them to begin to recognize nonprosodic
cues to syntactic phrases in the text. Complimentary gains in the prosodic
appropriateness of the child's oral reading and comprehension of the text
are predicted by Schreiber's theory explaining the effectiveness of repeated
reading.
Schreiber focused on the development of reading fluency in beginning
readers, but he also speculated (Schreiber, 1987) that poor readers may
have difficulty attaining reading fluency: poor readers, as with beginning
readers, continue to use prosody as a primary cue for chunking speech and
fail to recognize the need to make better use of other cues to phrasing in a
text. Indeed, in a study of phonological processing and comprehension
skills,
Mann, Cowin, and Schoenheimer (1989) demonstrated that poor
grade 4 readers' sensitivity to, and reliance on, prosodic cues resembled that
of younger (i.e., grade 2), reading ability matched children. Further, a
study conducted by Young and Bowers (in press) found that poor readers
in grade 5 read significantly less fluently and expressively than their aver-
age-ability peers, even when accuracy and rate were equivalent between the
groups. These same poor readers were also less able to segment text into
phrases. Finally, phrasal knowledge was shown to contribute to expressive-
ness and fluency in average, but not poor-ability, readers.
Designed to investigate the factors contributing to oral reading fluency
and expressiveness, the present study examined two components that are
common to many repeated reading methods. Specifically, prosodic model-
ing and practice with intact text were manipulated in order to determine
their effectiveness in producing gains in reading fluency, expressiveness,
and comprehension among poor readers.
While the discussion thus far has focused on examining reading perfor-
mance gains on trained text, the ultimate goal of any training method is to
facilitate the readers' ability to read previously unencountered text. Trans-
fer across texts following repeated reading training has been reported for
reading speed and fluency (Dahl, 1979; Dowhower, 1987; Faulkner
&
Levy,

Applied Psycholinguistics 17:1 62
Young et al.: Effects of modeling and repeated reading
1994;
Herman, 1985; Rashotte
&
Torgesen, 1985; Samuels, 1979; van Bon,
Boksebeld, Freide, & van den Hurk, 1991), reading comprehension (e.g.,
Conte & Humphreys, 1989; Dowhower, 1987), and word accuracy (e.g.,
Conte & Humphreys, 1989; Dahl, 1979; Dowhower, 1987). Two studies
reported in the literature (Faulkner & Levy, 1994; Rashotte & Torgesen,
1985) examined the nature of the representation involved in across-text
transfer for rereading practice. Both of these studies provide evidence that
the transfer of oral reading fluency gains can be accounted for by single-
word representations rather than higher order, text-level representations.
Herman (1985) drew a similar conclusion from a study in which she found
that, while reading rate and word accuracy transferred across passages, the
number of speech pauses remained relatively consistent. Herman concluded
that the observed improvements did not include improved phrasing, but
rather reflected gains in the speed and accuracy of individual word recogni-
tion.
All of the studies cited here examined transfer across passages following
rereading practice without a prosodic model. Dowhower's (1987) study
provides the only direct comparison discovered in the literature of across-
text transfer when a prosodic model either is or is not provided. Transfer of
reading speed, word accuracy, and comprehension gains similar to those
described here were observed in both repeated reading procedures. How-
ever, only a repeated reading practice that included a prosodic model facili-
tated the transfer of prosodic reading (i.e., decreases in pausal intrusions
and inappropriate phrases and an increase in the intonational marking of
phrases) to a similar but unpracticed text. These results suggest that reread-
ing practice with a prosodic model facilitates improvement beyond single-
word recognition skills and may include skills (e.g., parsing text into mean-
ingful phrases) that are reflected in a child's ability to read aloud with
expression. It is important to note, however, that the subjects in Dowhow-
er's study had average or better decoding skill, but relatively slow reading
rates.
It is unclear, therefore, whether prosodic modeling will facilitate the
transfer of prosodic reading in poor readers who have difficulty with decod-
ing as well as reading rate.
This study addressed the issue of transfer by comparing the effectiveness
of prosodic modeling versus text rereading in facilitating reading fluency
and expressiveness transfer across passages within the same text and across
readings of different texts. Dowhower's (1987) method of controlling dif-
ferences in passage characteristics (e.g., content, character, and readability)
was adopted: reading passages were divided in
half;
the first half was used
for training, and the second half was used for testing across related text
transfer effects. Two stories were divided in this manner and used as stimuli
in the study. As in Dowhower's study, the first half of a third, unpracticed
story was read as a pretest, and the other half was read as a posttraining
test to measure transfer across unrelated text.
To summarize, this study was designed to address the following ques-
tions:

Applied Psycholinguistics 17;.l . 63
Young et al.: Effects of modeling and repeated reading
1.
What are the independent and joint effects of prosodic modeling and text
repetition practice on the reading fluency and comprehension of poor
readers?
2.
Do the effects of prosodic modeling and text repetition practice transfer
to fluency, expressiveness, and comprehension of related or unrelated
text?
METHOD
Subjects
All subjects were drawn from a sample of 199 grade
5
children screened for
an earlier study that examined the determinants of reading fluency and
expressiveness (Young & Bowers, in press). A total of 40 children were
identified as poor readers according to the following selection criteria: a
standard score of 43 or less (24th percentile and below) on the Gates-
MacGinitie Reading Comprehension Test (MacGinitie
&
MacGinitie, 1978)
and at least one of the following characteristics: (1) teacher reports as below
average in either decoding and reading comprehension; (2) a score on the
Woodcock Reading Mastery Test-Revised (Woodcock, 1987) Word Identi-
fication subtest below the 25th percentile for grade 5. Of final sample of 40
poor readers (25 boys,
15
girls), 37 children scored below the 25th percentile
on word identification, and the remaining 3 scored within the low-average
range, but were nominated by their teachers as below grade level in both
reading decoding and comprehension. The children were selected from reg-
ular classrooms; however, most received special services (in-class or with-
drawal remedial services) for reading difficulties. Exclusion criteria for this
study included a teacher's judgment that a child's reading achievement
might be related to physical disability, lack of knowledge of English, or
emotional or behavioral difficulties.
Materials
A set of three prose passages with a stated grade
3
reading level were chosen
from published works as stimulus materials.' The stories were modified to
provide passages of
300
words with first and last halves of
150
words each.
Passage difficulty level was selected to correspond to the level at which
the readers could recognize most of the words, but were still encountering
some difficulty with fluency (i.e., reading rate and expressiveness). Passage
difficulty ranged from late grade 2 to grade 3 (mean readability was grade
3.0) as estimated by an extension of the Harris-Jacobson readability for-
mula (Harris
&
Jacobson, 1982) and the Fry (1977) readability index.
In keeping with Rashotte and Torgesen's (1985) demonstration of the
importance of word commonality among stories to the carryover of gains
resulting from the repeated reading method, the amount of word overlap
was measured across the first and last halves of each passage and between

2
3
38
41
37
47
.8
.02
2
52
51
34
40
.7
.83
3
55
51
31
37
.6
.3
.8
Applied Psychoiinguistics 17:1
, 64
Young
et
al.: Effects
of
modeling and repeated reading
Table 1. Readability
levels
of experimental
passages
and word
overlap
within
each
story
Story
1
Story 2 Story 3
Estimated grade level, Harris-Jacobson
Estimated grade level, Fry Index
No.
of unique words - first half of story
No.
of unique words - last half of story
No.
of shared words - first and last halves
°lo
of shared words - first and last halves
the passages. As expected, the first and last halves of the stories contained
many shared words (mean proportion of shared words on first vs. second
story halves = 41.6%). A description of each of the three passages, includ-
ing readability levels, and the ratios of unique and shared words in each
story half are presented in Table 1.
In determining the amount of word overlap, each shared word was coun-
ted only once, regardless of the number of times it was repeated within and
between stories. For example, a high-frequency, function word such as the
occurred several times in each of the stories, but was counted as only one
shared word. A total of 42 words were common to the first halves of at
least two of the stories. For example, the word you occurred in the first half
of all three stories, while the word ring was shared by two of the three
stories. Both of these words were counted in the total of 42 shared words.
Moreover, 60 words which occurred in the first halves of at least one of the
three stories were common to at least one of the second halves.
Experimental design and procedure
Subjects were tested individually for a single session of approximately 1
hour in a small, relatively distraction-free room within the elementary
school they attended. They were randomly assigned to one of four experi-
mental treatment groups (10 subjects per group). In order to determine if
the treatment groups were relatively matched in cognitive ability and lan-
guage skills, measures of nonverbal reasoning (Block Design subtest from
the Wechsler Intelligence Scale for Children-Revised [WISC-R]) and re-
ceptive vocabulary (PPVT-R) were administered. The treatment groups did
not differ on these measures or on any of the reading ability selection
measures.
The modeling and text practice components of the repeated reading
method were systematically varied in a 2 (practice
vs.
no practice of reading
continuous text) x 2 (modeling vs. no modeling of prosodic reading) fac-
torial design. In order to simplify discussion of these factors, they will be
hereafter referred to as the practice and modeling conditions, respectively.
The experimental conditions and specific type of training included in each
are as follows:

Applied Psycholinguistics 17:1
65
Young
et
al.: Effects
of
modeling
and
repeated reading
1.
No Prosodic Model - No Practice of Text Condition (repeated list read-
ing).
In order to eliminate the contextual component of practice with text
while still allowing equivalent amounts of practice on individual words,
the order of words contained in the training texts (i.e., first half of each
training story) was scrambled. Children read each of the scrambled word
lists orally three times. Words that were repeated during the text were
repeated an equivalent number of times on the list. Different random
word orderings were presented for each training trial. Immediate feed-
back was given to correct reading errors or omissions.
2.
Prosodic Model - No Practice of Text Condition (repeated listening).
The children listened to the experimenter read the passage aloud three
times.
Care was taken to read in a fluent, expressive manner so that a
good model of prosodic reading was provided.
3.
No Prosodic Model - Practice of Text Condition (unassisted repeated
reading). The children read the passages orally three times. An immediate
correction of reading errors or omissions was provided.
4.
Prosodic Model - Practice of Text Condition (assisted repeated reading).
The children and the experimenter read each of the passages orally (in
unison) three times. The experimenter read at a speed just slightly faster
than the child's own rate and in an appropriately expressive and intonated
manner. Through this read-along method, a model of expressive reading
was provided and errors or omissions were corrected.
The assignment of treatment conditions representing the two levels of
both practice and model are presented here:
Prosodic
Model
No Prosodic
Model
Text Practice
Assisted Repeated
Reading
Unassisted Repeated
Reading
No Text Practice
Listening
Word List
Reading
Following Dowhower's (1987) procedure,
one
story
was
randomly
se-
lected
to
serve
as an
overall assessment
of
generalized gains.
The
first half
of this story (initial test)
was
administered before
any
experimental treat-
ment. Without further practice with
it,
the entire story was presented
at the
completion
of the
experimental treatments.
The
readministered first half
constituted
the
final test, while
the
second half
of
this story represented
the across-story generalization test.
The
remaining
two
stories served
as
treatment stories. These stories were also divided
in half,
with
the
first half
serving
as a
pretest
and
final half
as a
posttest.
The
latter half-story
of
each treatment story served
as the
within-story transfer test.
The
order
of
presentation
for
each
of the two
treatment passages
was
counterbalanced
within each treatment group
to
control
for
possible story order effects.
All
reading passages
and
training materials (e.g., word lists) were presented
on
standard, SYi
by 11
inch (21'/i
by 28
centimeter) white paper, which
was
laminated
for
durability.

Applied Psycholinguistics 17:1
66
Young
et
al.: Effects
of
modeling and repeated reading
Table 2. Sequence of
experimental
tasks,
passages,
and
measures in
Experiment
2
TaskMeasures
Overall Assessment, Initial Test Sequence
Initial test
-
unassisted reading
of
Part
1
of
overall assessment passage
(OAP)
Retelling
-
child retells Part
1
of
overall
assessment story
Training Sequence'
Pretest
-
unassisted reading
of
Part
1
of training passage
Retelling
-
child retells Part
1
training
passage
Training
task
-
Part
1
is practiced
3
times by URR, ARR, list reading,
or
listening
Posttest
-
unassisted reading
of
Part
1
Within-story transfer
-
unassisted read-
ing
of
Part 2
of
training passage
Retelling (posttest
&
transfer)
-
child
retells entire story
Overall Assessment, Final Test Sequence
Final test
-
unassisted reading
of
Part
1
of OAP
Across-story
generalization
-
unassisted
reading
of
Part 2
of
OAP
Retelling
-
retelling
of
entire OAP
Fluency rating, reading speed, accuracy
Proportion
of
propositions recalled
Fluency rating, reading speed, accuracy
Proportion
of
propositions recalled
Fluency rating, reading speed, accuracy
Fluency rating, reading speed, accuracy
Proportion
of
recalled propositions
from Parts
1
and 2
Fluency rating, reading speed, accuracy
Fluency rating, reading speed, accuracy
Proportion
of
recalled propositions
from Parts
1
and 2
of
OAP
"Repeated twice, once per passage.
The general procedure was constant across treatment conditions, as out-
lined in Table 2. Note that comprehension was assessed after the initial test,
pretest, posttest, and overall assessment passage by the child's oral retelling
of what was read. After the pretest and oral retelling of the first half of a
treatment story, three training trials on that same material occurred. Three
trials were chosen in view of O'Shea, Sindelar, and O'Shea's (1987) finding
that, while poor readers' comprehension of text increased from one to
three rereadings, no additional gains were observed from three to seven
rereadings. Thus, three training trials appear to provide sufficient practice
for improvements to occur. The specific format of the passage (i.e., intact
text or list) and the training procedure were determined by the student's
assigned treatment group. Immediately after the training trials on each
story, the children read the entire passage and were asked to retell it.

Applied Psycholinguistics 17:1 67
Young et al.: Effects of modeling and repeated reading
No reading assistance, error correction, or feedback was provided to the
subjects during the pretest, posttest or transfer trials for the trained stories,
or during the initial, final, and generalization trials for the untrained sto-
ries.
Assistance was provided during the treatment phase of the session, as
outlined in the description of the specific training conditions. The sequence
of tasks included in the experimental session are summarized in Table 2.
After the reading and comprehension tasks were completed, the Block
Design subtest from the WISC-R was administered. Standard administra-
tion procedures, as outlined in the WISC-R manual, were used for this task.
Measures
Reading
rate.
The time taken to read each story section orally was mea-
sured, and reading times were converted to words per minute.
Reading
accuracy.
Corrected and uncorrected oral reading errors of omis-
sion, substitution, and insertion were recorded initially as the child read
each passage orally and were rechecked from the tape recordings of the oral
reading. An accuracy score represented the proportion of words in a pas-
sage that were read correctly (i.e., were not read as uncorrected errors).
Fluency
ratings.
Two trained raters scored the children's oral renditions of
each 150-word passage of text for-fluency and expressiveness using the
Fluency Scale developed by Allington and Brown (1979) and modified by
Allington (1983); this rating scale was also used by Rasinski (1985). In
order to ensure that the raters were blind to whether the oral reading repre-
sented the pretest or posttest condition for a given subject, audiotapes were
constructed in which these conditions were randomly ordered. Ratings were
made on the qualities of the oral reading, including phrasality, fluency, and
expression. For example, a fluency score of 1 was assigned to an oral
reading perceived as a "word-by-word reading." A score of
3
was given to a
reading that was "primarily in phrases (2-3 words in length) but sometimes
word by word: sometimes gives phrases inadequate stress in relation to
syntax." A rating of
6
was assigned to a performance summarized as "reads
in phrases, with fluency, using both terminal and internal punctuation;
provides appropriate semantic and syntactic emphasis for purposes of dra-
matization; expression approximates normal speech." Independent ratings
were made for each 50-word section of text. Fluency ratings for a particular
passage were determined by averaging the ratings made after each 50-word
section. Interrater reliability was established for overall fluency ratings by
randomly selecting 10% of the recorded oral readings. The agreement be-
tween raters, as indexed by the product-moment correlation, was .85. The
percentage agreement for fluency ratings made by the two raters was deter-
mined to be 94.4%.2
Story
retelling.
Reading comprehension was measured using an unstruc-
tured retelling of the story passages that were orally read. After reading the
first half of a story, the subjects were told: "Tell me everything you can

Applied Psycholinguistics 17:1
68
Young
et
al.: Effects
of
modeling and repeated reading
remember about the story, starting from the beginning." No probes or cues
were provided during or following the retelling to prevent these cues after
the pretest from influencing the posttest retellings. The posttest retelling
followed the subject's reading of an entire story. All of the story retellings
were tape-recorded and later transcribed.
The children's retellings of each story passage were scored according to
the recommendations for propositional analysis by Turner and Greene
(1978) and were based on Kintsch's propositional theory in
The
Representa-
tion of Meaning in Memory (1974). Each of the experimental passages were
analyzed to form a template text base, which contained all of the directly
expressed and inferred propositions contained in the text. This list of prop-
ositions is referred to as the "template" text base because it serves as a
model against which the children's retellings can be compared. The template
text bases were developed using the guidelines provided by Turner and
Greene (1978).
For the purpose of scoring, the propositions in each retelling were com-
pared with the propositions of the template text base. Credit was assigned
for every text proposition represented in the template text base that was
correctly recalled in the child's retelling. The scoring criterion allowed for
acceptable condensations of embedded propositions (such as "clothes" for
"shirt and pants") and for synonyms for the words contained in the original
text. Generalizations or abstractions of relations within a proposition were
not allowed as these would change the meaning of the proposition from
that represented in the text. A point was assigned for each correctly recalled
proposition. Thus, the retelling score represents the proportion of proposi-
tions contained in the template text base that the child recalled in the re-
telling.
Interrater reliability was established by randomly selecting 10% of the
retellings. Interrater reliability, as indexed by the correlation between the
total scores given by the two raters, was .96. The percent agreement, de-
fined as an assessment of each proposition as present or absent by the two
raters,
was 90%.
RESULTS
Two analytic strategies were used to answer the questions posed by this
study. The first applied an analysis of variance (ANOVA) approach to the
data to examine the treatment effects on reading performance. Preliminary
analyses, including descriptive analyses and transformations of data where
appropriate, are also described here.
Descriptive statistics
Table 3 presents a summary of the reading ability and individual difference
measures taken for the 40 children (divided into four treatment groups).
The means for the four treatment groups were below grade level for chil-
dren in grade 5 on all of the reading measures. These results are consistent

Applied Psycholinguistics 17:1
Young et
al.:
Effects of modeling and repeated reading
Table 3.
Characteristics
of
children in
four
treatment groups (means and
standard deviations)
69
Male
Female
Reading comprehension
(Gates t score for
grade 5)
Vocabulary knowledge
(PPVT-R)
Word identification skill
(Voile for grade 5)
Block Design subtest -
standard score
N
N
M
SD
M
SD
M
SD
M
SD
Repeated
list reading
6
4
39.2
2.5
90.3
7.5
7.9
9.5
9.7
2.8
Repeated
listening
7
3
37.3
4.7
91.0
12.9
14.0
14.1
8.3
2.5
Unassisted
repeated
reading
5
5
36.0
3.2
97.6
10.2
7.5
9.7
8.9
3.9
Assisted
repeated
reading
7
3
39.1
3.6
90.7
13.6
10.7
12.1
10.6
4.3
with the expectations derived from a sample of exclusively poor readers.
However, the treatment groups did not differ significantly on any of the
reading ability or individual difference measures. Thus, no adjustments
were necessary in subsequent analyses for treatment condition effects.
Data transformations
An examination of the range and distribution of the proportion of correctly
read words revealed that the distribution of scores was limited. In particu-
lar, the mean accuracy scores across passages and treatment conditions was
approximately 95% correct. Thus, an arcsine or angular transformation
was performed on the data such that the proportions near
1
(100% correct)
could be spread out and the variance increased. This transformation re-
sulted in a data set more suited to the analysis of variance procedure, given
its assumptions of homogeneity of variance and normality.
While the story-retelling measure is also proportional (i.e., the propor-
tion of correctly recalled propositions contained in a text), an examination
of these data revealed a more normal distribution and broader range of
scores than in the accuracy data. In addition, preliminary analyses revealed
roughly identical results when the data was transformed versus untrans-
formed. For ease of interpretation, therefore, the story-retelling results
described here are based on analyses of the untransformed proportions.
Passage order effects
One area of concern in any study in which a repeated measures design is
used is the possibility of carry-over effects, which may result from the
ordering of the conditions. In the present study, the order with which

Applied Psycholinguistics 17:1
70
Young
et
al.: Effects
of
modeling and repeated reading
the two treatment passages were presented was counterbalanced to control
potential carry-over effects. Despite this control, the data were examined
for potential carry-over as a preliminary step to combining each subject's
performance on the two treatment passages into a single score. No consis-
tent or problematic carry-over effect of passage order in the performance
of the subjects across the treatment condition groups was found. Thus,
each subject's performance on the two treatment passages could be mean-
ingfully combined, and this single score is included in the analyses reported
here.
The means and standard deviations for the treatment groups on the
pretest, posttest, and transfer tests are presented in Table 4 in terms of the
two levels of text practice (practice of text vs. no practice of text) and
prosodic modeling (model of prosody vs. no model of prosody).
Intercorrelation among variables
The intercorrelations among study variables are presented in Table 5. The
PPVT-R and Block Design scores did not correlate significantly with any of
the reading behavior or other individual difference variables. This indicates
that neither vocabulary knowledge nor nonverbal reasoning ability pre-
dicted reading performance within this sample of exclusively poor readers.
Therefore, neither of these variables appears in Table 5. As expected, most
of the reading performance measures were highly intercorrelated. This was
particularly true of fluency and reading speed, which were strongly corre-
lated across all testing periods. Comprehension maintained the strongest
relationship with fluency, while being more moderately related to both
accuracy and reading speed.
A comparison of the intercorrelations between accuracy and the other
reading measures revealed moderate to strong correlations on the pretest
and transfer test periods, but no significant correlations with any of the
other measures of reading skill on the posttest. This finding may reflect the
uniformly beneficial effects of error correction provided during the training
trials on word accuracy. Consistent with this hypothesis, the intercorrela-
tions between accuracy and the other reading measures on the final test of
the overall assessment passage (i.e., the story on which no error correction
was provided) did not show the same decline; again, accuracy reflected
general reading ability rather than the error correction procedure.
Treatment group effects
The primary analytic approach taken in this study was a mixed-model
ANOVA which compared the effects of practice of text versus prosodic
modeling on the children's pretest versus posttest and pretest versus transfer
test reading performances. Thus, a 2 (text practice) x 2 (prosodic model-
ing) x 2 (test, i.e., pretest and posttest or pretest and transfer test)
ANOVA, with repeated measures on the last factor, was applied to the
data. The treatment condition by test interaction represents the effect of

Table 4. Means and standard deviations of
the practice
of text and
prosodic
modeling conditions for
within-story
test
trials
on
reading
rate,
fluency
ratings,
word
accuracy,
and comprehension
Practice of text
(unassisted and
assisted repeated
reading), N = 20
No practice of text
(repeated list and
repeated listening),
N = 20
Model of prosody
(listening and
assisted repeated
reading), N = 20
No model of prosody
(repeated list and
unassisted repeated
reading), N = 20
M
SD
M
SD
M
SD
M
SD
Pre.
103.9
31
113.4
37
110.7
33
106.7
37
Reading
Post.
140.8
33
131.1
36
136.2
37
135.7
33
rate
Transfer
104.5
28
107.3
35
106.3
31
105.5
32
Pre.
4.2
.72
4.4
.69
4.3
.71
4.3
.71
Fluency
Post.
4.8
.72
4!6
.65
4.7
.73
4.7
.66
Transfer
4.4
.55
4.5
.62
4.4
.58
4.4
.60
Word accuracy
Pre.
95.6
3.5
96.8
2.4
96.0
3.6
96.4
2.5
Post.
97.9
1.7
97.5
1.5
97.4
1.9
97.9
1.2
Transfer
95.5
3.6
96.0
2.5
96.2
2.9
95.3
3.3
Comprehension
Pre.
33.3
9.5
32.0
11.7
29.0
10.7
36.2
9.2
Post.
43.4
12.1
36.1
14.4
36.3
15.2
43.2
11.3
Transfer
36.9
11.9
37.8
14.4
34.3
12.8
40.3
12.8

Table 5. Correlation among
variables
contributing to
reading
rate,
fluency, and
accuracy
in the
practice passages
Pretest Wpm
Pretest Fluen.
Pretest Accur.
Pretest Comp.
Posttest Wpm
Posttest Fluen.
Posttest Accur.
Posttest Comp.
Trans.
Wpm
Trans.
Fluen.
Trans.
Accur.
Trans.
Comp.
Pretest
Wpm Fluen. Accur.
X .86 .29
X .52
X
Comp.
.22T
.43
.31
X
Wpm
.89
.79
n.s.
.34
X
Posttest
Fluen.
.31
.80
.29
.39
.82
X
Accur.
•23t
.33
.64
n.s.
n.s.
n.s.
X
Comp.
n.s.
.37
.28
.86
.32
.41
• 25t
X
Wpm
.95
.83
•25t
.30
.91
.74
n.s.
• 24t
X
Transfer test
Fluen.
.85
.88
.31
.41
.85
.84
n.s.
.34
.89
X
Accur.
.52
.67
.77
n.s.
.46
.51
.47
n.s.
.49
.58
X
Comp.
.29
.53
.42
.80
.35
.40
n.s.
.69
.36
.52
.35
X
Note: Wpm = words per minute; fluen. = fluency; accur. = accuracy; comp. = comprehension; trans. = transfer. All
correlations listed are significant at p < .05 or above except tp < .10.

Applied Psycholinguistics 17:1 73
Young et al.: Effects of modeling and repeated reading
practice with text or prosodic modeling for a specific comparison. For
example, the effect of prosodic modeling on transfer within a story is evi-
dent in the prosodic model by test (pretest vs. transfer test) interaction.
Pretest versus posttest training effects
On the training passages, univariate tests revealed that overall gains from
pretest to posttest trials were significant for fluency, F{\, 36) = 40.4,
p < .001; reading speed, F(l, 36) = 166, p < .001; accuracy, F(l, 36)
= 18.5,/? < .001; and comprehension, F(l, 36) = 46.9, p <
.001.
These
results indicate that, on average, the children's reading performance im-
proved on the story halves on which they were trained.
The practice of text versus no practice of text comparison was examined
in the Practice x Test interaction. This interaction was significant for flu-
ency, F(l, 36) = 5.67, p < .05; reading speed, F(l, 36) = 20.72, p <
.001;
word accuracy, F(l, 36) = 4.66, p < .05; and comprehension, F(l,
36) = 8.39, p <
.01.
These results indicate that the children who practiced
the text in meaningful context (practice of text condition) showed signifi-
cant improvement over those who did not (no practice of text condition).
The modeling of prosodic reading versus no modeling of prosodic read-
ing comparison yielded no significant differences in fluency, reading rate,
accuracy, or comprehension. Thus,.the modeling of prosodic reading did
not differentially add to the improvements in reading from pretest to post-
test. In addition, no effect of a Practice x Model x Time interaction was
found for any of the reading performance measures. This lack of a signifi-
cant interaction indicates that the training groups within the practice and
modeling conditions varied consistently. In particular, no single training
procedure (e.g., assisted repeated reading) can account for the model and
practice condition results.
Within-story transfer effects
Univariate tests examining the differences between the first reading of the
training passages and first reading of the transfer passages (i.e., later half
of the training passages) indicated significant improvements in fluency,
F(l,
36) = 7.33, p < .01, and comprehension, F(l, 36) = 13.58, p <
.001,
which did not interact with the text practice or prosodic modeling
conditions. Thus, regardless of the type of training received, the children
read with greater fluency and expressiveness and recalled more propositions
from the second half of the treatment stories (after training) than they
did on their initial reading of the first half of these same stories (before
training).
An examination of within-story transfer for reading speed revealed that
only the text practice effect was significant, F(l, 36) = 4.21, p < .05.
In particular, subjects who practiced with intact text (repeated reading)
maintained the same reading speed on the transfer passage as they did on
the initial reading of the training passage. In contrast, paired t tests con-

Applied Psycholinguistics 17:1
74
Young
et
al.: Effects
of
modeling and repeated reading
firmed that reading rates for subjects in the no text practice condition were
significantly slower on the transfer passage than pretest, r(19) = 4.9, p <
.001.
Finally, a significant Practice x Model x Test interaction was
found for word accuracy, F(l, 36) =
8.20,p
<
.01.
Paired t tests revealed
that only assisted repeated reading practice improved accuracy on the trans-
fer test, /(9) =
—3.9,
p < .01, while a trend toward less accurate transfer
test performance was revealed among all other training conditions.
To summarize, the within-story transfer of fluency ratings and compre-
hension was facilitated by training, regardless of whether this training in-
cluded practice with intact text or prosodic modeling. Reading time on the
transfer passage decreased in the no text practice condition, but remained
relatively stable among subjects who received practice with intact text. Only
assisted repeated reading (i.e., practice that included both repeated reading
and prosodic modeling) facilitated the transfer of word accuracy.
Across-story transfer effects
The final set of comparisons between the treatment conditions included
comparing the initial versus the final test, as well as the initial versus the
generalization test, of the overall assessment story. As noted, this story was
slightly more difficult (as estimated by readability indices) than either of
the two training stories. Table 6 presents the means and standard deviations
for the text practice and prosodic modeling conditions across these test
trials.
Recall that this story was not included in the training trials, but
instead was used to assess across-story transfer of the treatment condition
effects. Thus, the children never received experimenter assistance or feed-
back regarding reading errors on this story.
Univariate tests of the initial versus final tests of the overall assessment
passage revealed the following results. Significant improvements were
noted in reading rate, F{1, 36) = 5.55, p < .05, and comprehension,
F(\, 36) = 25.92, p < .001, on the final test over the initial test. On aver-
age,
the children reread the first half of the overall assessment passage with
greater speed and recalled more propositions during retelling after training
on the other passages. There was a significant interaction, F{\, 36) = 7.4,
p < .01, between the text practice condition and the test for word accu-
racy. Paired t tests showed that final test word accuracy was significantly
higher than initial test accuracy only when the subjects practiced with intact
text. There was no difference between initial and final test accuracy among
subjects who did not practice with intact text. Finally, there was no differ-
ence in fluency ratings on the initial and final tests for any of the practice
conditions (i.e., text practice and prosodic modeling).
The across-story generalization effect was determined by comparing the
children's reading performance on the first and final halves of the overall
assessment passage. The initial test preceded training trials on the other
stories, while the generalization test occurred after the completion of these
trials.
A significant main effect for test (initial vs. generalization) was re-
vealed for all of the reading measures, including reading rate, F(l, 36) =

Table 6. Means and standard deviations for the practice of text and prosodic modeling conditions across story transfer trials on
reading rate, fluency ratings, word
accuracy,
and comprehension
Practice of text (unassisted
and assisted repeated
reading), N = 20
No practice of text (repeated
list and repeated listening),
N = 20
Model of prosody (listening
and assisted repeated read-
ing),
N = 20
No model of prosody (re-
peated list and unassisted re-
peated reading), N = 20
M
SD
M
SD
M
SD
M
SD
Reading rate
Initial
108.4
35
114.8
39
114.6
32.8
108.7
40.3
Final
115.0
32
119.5
40
119.9
36.5
114.6
35.5
Gen.
89.4
29
94.5
32
94.6
27.9
89.3
32.3
Initial
4.6
.75
4.7
.63
4.7
.67
4.6
.72
Fluency
Final
4.7
.59
4.8
.64
4.8
.65
4.7
.57
Gen.
4.4
.68
4.6
.51
4.5
.59
4.5
.62
Word
Initial
95.1
4.0
96.3
4.0
95.5
4.6
95.9
3.5
accuracy
Final
96.4
3.4
96.0
3.1
96.3
3.4
96.0
3.1
Gen.
93.7
5.9
94.7
4.1
94.4
4.8
94.0
5.4
Comprehension
Initial
23.4
10.6
22.7
12.1
21.5
11.6
24.7
10.9
Final
31.1
17.0
35.9
15.0
31.5
18.3
35.5
13.4
Gen.
33.2
12.3
30.3
16.4
30.7
16.9
32.7
11.7

Applied Psycholinguistics 17:1
76
Young
et
al.: Effects
of
modeling
and
repeated reading
72.9,
p < .001;
fluency,
F(l, 36) = 10.2, p < .01; and
word accuracy,
F(l,
36) = 11.3, p < .01.
This effect
was due to an
overall decline
in
performance
on the
final half
of the
overall assessment passage.
In con-
trast,
the
significant main effect
for
comprehension,
F(l, 36) =
23.0,
p <
.001,
reflected
an
increase
in the
number
of
propositions recalled during
the
generalization test.
No
significant effects were observed
for
either
of the
practice
or
modeling conditions. These results indicate that
the
second half
of
the
overall assessment passage
was
more difficult
for the
subjects than
the first
half.
Nevertheless,
the
children recalled more
of the
propositions
contained
in the
second half
of the
overall assessment story than those
in
the initial half when
it
was first encountered. This finding
is
understandable
when
the
amount
of
overlap between story halves
is
considered. Thus, even
though
the
final half
of the
story
was
read
for the
first time during
the
generalization test,
the
story characters, setting, theme,
and
many
of
the individual words
had
already been encountered
in the
first half
of the
story. This overlap
can
reasonably
be
expected
to
account
for the
improved
comprehension
on the
second half
of
the story.
Summary
Regardless
of the
specific type
of
treatment received, reading speed, accu-
racy, fluency,
and
comprehension improved after training
on the
training
passages.
In
addition, those children who received repeated reading training
showed significant additional gains
on all
reading performance measures
over those
who did not
practice with intact text. Modeling
of
prosodic
reading
did not
differentially improve reading performance
on
posttest
after training
or on any of
the transfer tests summarized here.
An examination
of
within-story transfer effects revealed improvements
in fluency
and
comprehension across treatment groups.
On
average, train-
ing
on the
first half
of a
story improved
the
children's comprehension
and
the fluency ratings they received
for
their initial reading
of the
second half
of
the
same story. Despite these gains
in
fluency
and
comprehension, only
those children
who
received repeated reading training (i.e., practice with
intact text) read
the
second half
of the
treatment stories
at
approximately
the same speed
as
they initially read
the
first
half.
Those
who did not
practice with intact text read these story halves slower. Finally, while word
accuracy declined slightly
on the
second half
of the
treatment stories
in
most conditions, those children
who
practiced using
the
assisted repeated
reading method showed improved word accuracy.
On average,
the
children reread
the
overall assessment story faster
and
recalled more propositions from
it
after training
on the
other stories than
they
did
when reading
the
same passage
for the
first time. Only those
children
who
received repeated reading practice
on the
training stories
showed improved word accuracy
on the
untrained, overall assessment
story.
A
significant decline
in
reading speed, fluency,
and
accuracy
was
observed across
all
treatment groups
on the
final half
of the
overall assess-

Applied Psycholinguistics 17:1
77
Young
et
al.: Effects
of
modeling and repeated reading
ment story.
In
contrast, comprehension
on
this later half
of
the story was
significantly higher than was demonstrated on the initial reading
of
the first
half
of
the same story.
DISCUSSION
Training effects within the same text
As expected, the training
of a
story passage significantly improved
a
child's
reading performance
on
that same passage.
In
general,
the
children read
texts significantly faster
and
recalled more propositions after they trained
on them, regardless
of
the particular type
of
training they received. While
a
significant main effect
for
training was also revealed
for
fluency and word
accuracy,
the
averaged gains across
all of the
treatment conditions were
considerably smaller than those
for
speed and comprehension.
Poor readers
in
the current study who practiced with intact text showed
greater gains in all aspects
of
their reading behavior than those who did not
practice with intact text. These results
are
consistent with those
of
several
other studies
in
which repeated reading
of
continuous text
was
shown
to
increase reading speed, accuracy, expressiveness, and comprehension.
Prosodic modeling was
not
associated with
any
differential reading
per-
formance
or
comprehension gains. Thus, Schreiber's (1980, 1987) conten-
tion that poor readers may read with greater fluency
and
comprehend text
better
if
they
are
provided with
a
good prosodic model was
not
supported
in
the
current study. Clearly,
the
most effective condition
for
improving
poor readers' speed, accuracy, expressiveness,
and
comprehension was
re-
peated reading
of
meaningful text. This finding speaks
to the
importance
of separating
the
characteristics
of
skilled reading performance from
the
methods used
to
promote such performance
in
less skilled readers.
In par-
ticular, the oral reading
of
poor readers who practiced with intact text was
more appropriately expressive,
as
indicated
by
perceptual ratings
of flu-
ency. Nevertheless, the modeling
of
expressive reading was
not
responsible
for improvements in this aspect
of
the children's reading.
The pattern
of
results from
the
current study
can be
compared with
relevant research findings
to
determine whether
the
gains associated with
rereading resulted from word identification
or
higher order reading skill.
The question
of
particular interest
in
this regard is, why were poor readers'
comprehension and reading expressiveness improved more by rereading text
than by other forms
of
training? Some researchers, most notably Schreiber
(1980)
and
Dowhower (1987), interpreted
the
improvements
in
prosodic
reading
and
comprehension gained from
a
rereading practice
as
evidence
that
the
children went from single-word
to
phrasal reading. Thus,
Dowhower (1987) commented that repeated reading
of
meaningful text
"helped children tacitly develop prosodic strategies
for
organizing text"
(p.
403).
While the prosodic characteristics
of
oral reading in the current study
were
not
measured
in
terms
of
acoustic properties
as
they were
by
Dowhower, the average rating
of
fluency on the pretest reflected
a
monoto-

Applied Psycholinguistics 17:1
78
Young
et
al.: Effects
of
modeling and repeated reading
nal but mostly phrasal reading style, while the average posttest rating was
indicative of a more naturally expressive and an entirely phrasal reading. If
we accept Dowhower and Schreiber's interpretation of similar shifts in oral
reading behavior, then we can point to the more appropriate phrasing and
expression on posttest as evidence of the children's movement into a higher
order (i.e., phrasal level) processing of text. According to this account, the
prosody provided by the model in the assisted rereading condition and the
child's own use of prosody in the unassisted rereading condition serve as a
linguistic guidepost, directing the poor readers' attention to syntactic phras-
ing in the text they might otherwise miss. Thus, the text is processed at a
higher level, and comprehension is enhanced.
While this account is compelling and intuitively attractive, it is hard to
reconcile it with certain patterns of results within the current study or with
findings cited in the research literature. Contrary to the predictions based
on this explanation, there was the lack of any differential improvement in
fluency and comprehension when a prosodic model was provided. Simi-
larly, few significant differences were reported by Dowhower (1986), the
only other study in the literature that directly compared repeated reading
with and without a prosodic model. An interesting finding of the current
study is that text rereading resulted in somewhat larger gains in comprehen-
sion than did listening to the text being read.3 This finding was unexpected
because listening comprehension is commonly found to be superior to read-
ing comprehension among poor readers. Swalm (1972), for example, re-
ported that below-average readers comprehend best after listening, next
best after oral reading, and worst after silent reading. Many informal read-
ing inventories use a child's listening comprehension as an estimate of the
potential to comprehend text without the limitations imposed by weak de-
coding skills.
Even if reading and listening comprehension do not reflect exactly the
same processes (Carlisle & Felbinger, 1991), few would dispute that com-
prehension gains from listening are not confined to word meanings, but
include higher level representations of text meaning as well. Given that the
training passage was read to the children in the listening condition three
times,
it is reasonable to assume that they had at least as many opportuni-
ties to develop representations of the propositions contained in the text as
those children who read for themselves. Further, the children's demeanor
during the listening trials and their comments at the completion of the study
suggest that they were neither bored nor inattentive while listening to the
story readings.
Since comprehension increases were slightly greater in the repeated read-
ing than in the listening groups, it appears unlikely that all of the gains
associated with rereading result from the readers' increased awareness of
syntactic boundaries or higher level processing of text. An alternative hy-
pothesis is that increases on single-word identification skills are associated
with gains in fluency and comprehension. This hypothesis is consistent with
the findings of Rashotte and Torgesen (1985) and Faulkner and Levy
(1994);
these studies found that gains in reading speed from rereading

Applied Psycholinguistics 17:1
79
Young
et
al.: Effects
of
modeling and repeated reading
among poor readers varied with
the
degree
of
word overlap among
the
stories. Faulkner
and
Levy demonstrated that these gains cannot
be
attrib-
uted to the more general contextual similarities among the stories.
But how can this position be reconciled with the results
of
this and other
studies (e.g., Dahl, 1979; Fleisher, Jenkins,
&
Pany, 1979; Spring, Blunden,
& Gatheral, 1981) that show that training
in
rapid decoding
for
individual
words does
not
transfer
to the
comprehension
of
intact text?4 Stanovich
(1986) resolved much
of
the confusion regarding context effects
by
distin-
guishing
the
effects
on
word recognition from those most relevant
to
text
comprehension. His review
of
numerous studies revealed that poorer read-
ers rely
on
context even more than skilled readers when
the
task
is
word
recognition, while skilled readers rely more
on
context
in
comprehension
tasks.
The poor readers
in
the current study, therefore, may have been able
to identify the individual words faster and easier when they were embedded
in
a
meaningful context.
In
keeping with Perfetti's (1985) verbal efficiency
theory, gains
in the
speed
and
efficiency
of
word recognition would allow
for shifts
in
processing resources
to
other aspects
of
reading such
as
com-
prehension.
In
this regard,
the
accompanying gains
in
fluency
and
expres-
siveness
are
by-products
of
word recognition efficiency
and
text compre-
hension.
Training transfer effects across related text
The results described here demonstrate that practice with intact text
im-
proves
the
oral reading performance
and
the comprehension
of
poor read-
ers.
Of at
least equal importance, however,
is the
question,
do
these gains
generalize
or
transfer
to
untrained passages? Transfer was assessed
by ex-
amining the effects
of
textual practice and prosodic modeling
on
texts that
were highly similar
to the
two training texts. Given that transfer was mea-
sured
on the
second half
of the
training passages,
the
similarities include
overlaps
in
words, theme, characters, and setting. This high degree
of
over-
lap was used
to
ensure that the passages were equated
in
terms
of
readabil-
ity, vocabulary level,
and
story structure, while maximizing
the
likelihood
of detecting any transfer that resulted from the training session.
On average,
all of
the training conditions produced transfer
in
terms
of
an improved rating
of
oral reading fluency and the number
of
propositions
recalled after training. The additional gains
in
fluency and comprehension,
which were seen on the training texts due to the repeated reading conditions
(text practice),
did not
transfer
to
highly similar texts. Only children
in the
repeated reading conditions showed
a
transfer
of
reading speed gains across
texts.
Finally, children who received assisted repeated reading training read
the transfer text with greater accuracy than they initially read
the
training
passage.
A
decline
in
accuracy was noted
on
the transfer text
for all of
the
other training conditions. The presence
or
absence
of a
prosodic model
did
not significantly influence oral reading behavior
or
comprehension
on the
untrained text.

Applied Psycholinguistics 17:1
80
Young
et
al.: Effects
of
modeling and repeated reading
Training generalization effects across nonrelated text
Recall that the first half of an untrained passage was given before any
treatment, and that both the first and second halves of this story were read
after the training on different texts was completed. Significant improve-
ments across training conditions in reading rate and comprehension were
noted on the second reading of the first half of the untrained text. Children
who practiced with intact text also showed significantly improved word
accuracy over those who did not. Finally, only improvements in compre-
hension transferred to the last half of this story. All other measures of
reading performance (including reading rate, accuracy, and fluency) actu-
ally declined on this text.
The improvement in comprehension across all conditions, even when
other measures of reading performance actually declined (i.e., on transfer
of the final half of the overall assessment passage), suggests that reading
the first half of each passage and then retelling it for the purpose of pretest
measurement may, in
itself,
have enhanced comprehension on posttest.
This interpretation of the general gain in comprehension across conditions
appears quite reasonable in view of findings reported by Gambrell, Koski-
nen, and Kapinus (1991) in which story-retelling practice was shown to
improve reading comprehension performance in both proficient and less-
proficient readers.
Theoretical implications
This study examined Schreiber's (1980) contention that the provision of
a prosodic model is an important component of reading practice, which
contributes specifically to the development of fluent, phrasal reading in
those children who read haltingly. The present study provides clear support
for the facilitating effect of text practice, but little support for the conten-
tion that prosodic modeling contributes to reading gains among poor read-
ers.
Thus, rereading of the same text facilitated the expressiveness of read-
ing and use of appropriate phrasing (as suggested by improved ratings on
the fluency scale). The transfer of rereading gains to a novel text that was
highly similar to the training text was confined to reading speed and, for
the assisted repeated reading group only, accuracy. Rereading gains in accu-
racy also transferred to a text that was contextually dissimilar to the treat-
ment texts.
An important issue to be addressed when discussing these results is the
role of feedback provided by the experimenter in each experimental condi-
tion. In order that the groups be equated for feedback, on-line correction
was provided for misread or omitted words in all of the reading conditions.
It is possible that this feedback emphasized accuracy rather than fluency or
comprehension. Thus, the pattern of results may have been, at least in part,
produced by the emphasis on accuracy and the interruption of "higher
order skills" involved in fluency and comprehension.5 However, it could be
argued that the lack of difference between training with unassisted repeated

Applied Psycholinguistics 17:1 81
Young et al.: Effects of modeling and repeated reading
reading (where continuous reading was interrupted when misread words
were corrected) and the assisted repeated reading (where no interruption
of reading flow occurred) speaks against this interpretation. Further, the
relatively weak gains in the listening condition, during which no interrup-
tion for correction occurred, offers little support for feedback as a differen-
tial determinant of the observed results.
Instead, many of the gains associated with rereading intact text can be
explained by the contextual facilitation of several levels of text processing.
Adams's (1990) discussion of the interaction between several processes (i.e.,
orthographic, phonological, and semantic) involved in reading acquisition
and skilled reading provides the basis for interpreting the current results.
Adams described context as an aid to decoding when the reader is slow to
recognize words that are within his or her general reading ability level. Such
slow recognition suggests that the available orthographic information, by
itself,
is not sufficient for identification. In skilled readers, this effect is
only apparent when orthographic processing is interrupted, as in the de-
graded or very quickly presented stimuli in laboratory experiments. For
unskilled readers with limited knowledge of orthography and generally
poor decoding skills, context can assist word recognition in normal text
conditions.
Adams (1990) explained this relationship as follows: "First, where a spell-
ing pattern is only partially processed, contextual excitation can augment
orthographic excitation so as to select the intended word from any competi-
tors.
Second, where orthographic processing is laborious or uncertain, indi-
rect excitation, originating in the Context processor, can help to speed its
progress" (p. 140). This view is consistent with Stanovich's (1980) interac-
tive compensatory model of model of word identification. In the current
study, the practice of words in scrambled format (i.e., no meaningful con-
text) resulted in consistently smaller gains than when the same words were
reread in a meaningful context. Some authors have concluded from similar
results that poor readers have difficulty transferring single-word skills to
text (e.g., Fleisher et al., 1979). An alternative explanation is that, for
readers with limited word recognition skills, gains in decoding ease and
speed are provided by automatic contextual facilitation. Given that the level
of text difficulty was appropriate for the children in this study and that all
of the training conditions provided information that was relevant to the
accuracy of decoding, practice with orthographic representations of words
while repeating the context provided conditions that were ripe for the con-
textual facilitation of orthographic skill, as shown especially by increased
reading speed and expressiveness.
The preceding discussion of the importance of word recognition skills is
not intended to imply that higher level comprehension skills are irrelevant.
Text-phrasing ability, for example, is often cited as an important factor in
skilled reading, particularly in fluency and comprehension. In his study of
factors that contribute to reading comprehension in average-ability grade 3
and 5 readers, Rasinski (1985) determined that text phrasing appears to be
dependent on automatic (text-driven) and contextual (concept-driven)

Applied Psycholinguistics 17:1
82
Young et al.: Effects
of
modeling and repeated reading
word identification skills. Thus, although text phrasing or syntactic sensitiv-
ity are factors that influence reading fluency and comprehension, word
identification skills, especially in young or poor readers, remain an impor-
tant influence. The greater number of propositions recalled after practice
by rereading text may have been, at least in part, due to the facilitation of
word recognition provided by context, which in turn allowed processing
resources to be allocated to comprehension.
Taken together, the findings of the current research and a review of
relevant literature suggest the following. While fluency differences between
poor and average readers may, in part, reflect differences in the degree to
which they apply phrasal knowledge to reading, the remedy may not lie in
targeting phrasing ability as a main teaching strategy. This conclusion is in
keeping with the findings of Taylor, Wade, and Yekovich (1985), who
compared the rereading practice of standard text with the effectiveness of
an intervention designed to help grade 5 students recognize phrasal bound-
aries by providing them with a text in which the phrasal units were sepa-
rated on the page. The practice of text rather than phrasing was found to
facilitate poor readers' recall of idea units in the text.
Prosodic modeling may not have been helpful to poor readers in the
present study because this intervention did not also provide a. sufficient
means of improving word recognition skills. Rather, repeated reading prac-
tice with meaningful text enables contextual facilitation of weak decoding
skills as well as opportunities to group text in meaningful units and thus
appears to be most advantageous.
ACKNOWLEDGMENTS
This research was completed as part of a doctoral dissertation by Arlene Young,
University of Waterloo, Canada.
NOTES
1.
The story titles and their sources were as follows: Reading Mastery III, Text-
book A (Chicago: Science Research Associates, 1988). "Don Makes The Fish
Too Hungry," Lesson 8, pp. 25-27; "Nancy Wants To Stay Little," Lesson 32,
pp.
127-130. Across the
World,
an anthology compiled and edited by Zena
Sutherland and Marilyn Cunningham (Illinois: Open Court Publishing Com-
pany, 1989). "How the Firefly Got Its Light," pp. 162-163.
2.
Given that the fluency scale is an ordinal scale with intervals of
1
point, agree-
ments included ratings that varied by less than 1 point. Thus, if one rater
assigned a fluency rating of 4.3 for a reading sample while the other rater
assigned a rating of 4.0, this was recorded as an agreement.
3.
A single degree of freedom contrast comparing pre- and posttest change in
comprehension for the listening group versus the combined repeated reading
groups approached significance, F{\, 36) = 3.19, p < .08. Comparisons be-
tween comprehension gains made by the unassisted repeated reading group (in
which no reading model was provided) and the listening group also approached
significance, F(l, 36) =
3.73,
p < .06.

Applied Psycholinguistics 17:1 83
Young et al.: Effects of modeling and repeated reading
4.
A single degree of freedom contrast comparing the gains made by the word list
practice group and the combined repeated reading groups confirmed that the
gains were smaller when the words were practiced out of context, F(l, 36) =
8.67, p < .01.
5.
The authors thank reviewer
B
for bringing this interpretation to their attention.
REFERENCES
Adams, M. J. (1990). Beginning to
read:
Thinking and learning about print. Cambridge, MA:
MIT Press.
Allington, R. L. (1983). Fluency: The neglected reading goal. Reading Teacher, 36, 556-561.
Allington, R. L., & Brown, S. (1979). Fact: A multi-media reading program. Milwaukee, WI:
Raintree Publishers.
Bowers, P. G. (1993). Text reading and rereading: Predictors of fluency beyond word recogni-
tion.
Journal of Reading Behavior, 25, 133-153.
Carlisle, J. F., & Felbinger, L. (1991). Profiles of listening and reading comprehension.
Journal of Educational Research, 84, 345-354.
Conte, R., & Humphreys, R. (1989). Repeated readings using audiotaped material enhances
oral reading in children with reading difficulties. Journal of Communicative Disorders.
22,
65-79.
Dahl, P. R. (1979). An experimental program for teaching high-speed word recognition and
comprehension skills. In J. E. Button, T. C. Lovitt,
&
T. D. Rowland (Eds.), Commu-
nications research in learning disabilities and mental retardation (pp. 33-65). Balti-
more, MD: University Park Press.
Dowhower, S. L. (1987). Effects of repeated reading on second-grade transitional readers'
fluency and comprehension. Reading Research Quarterly, 22, 389-406.
Downs, J., & Morin, S. (1990). Improving reading fluency with precision teaching. Teaching
Exceptional Children, Spring, 38-40.
Faulkner, H. J., & Levy, B. A. (1994). How text difficulty and reader skill interact to produce
differential reliance on word and content overlap in reading transfer. Journal of Experi-
mental Child Psychology, 50, 1-24.
Fleisher, L. S., Jenkins, J. R., & Pany, D. (1979). Effects on poor readers' comprehension of
training in rapid decoding. Reading Research Quarterly, 15, 30-48.
Fry, E. (1977). Fry's readability graph: Clarifications, validity, and extension to level 17.
Journal of Reading, 21, 242-252.
Gambrell, L. B., Koskinen, P. S., & Kapinus, B. A. (1991). Retelling and the reading compre-
hension of proficient and less-proficient readers. Journal of Educational Research, 84,
356-362.
Harris, A. A., & Jacobson, M. D. (1982). Basic reading vocabularies (pp. 19-37). New York:
Macmillan.
Heckelman, R. G. (1969). A neurological impress method of reading instruction. Academic
Therapy, 4, 277-282.
Herman, P. A. (1985). The effect of repeated readings on reading rate, speed pauses, and
word recognition. Reading Research Quarterly, 20, 553-565.
Kintsch, W. (1974). The representation of meaning in memory. Hillsdale, NJ: Erlbaum.
Krug, D., Davis, B., & Glover, J. (1990). Massed versus distributed repeated reading: A case
of forgetting helping recall? Journal of Educational Psychology, 82, 366-371.
LaBerge, D., & Samuels, S. J. (1974). Toward a theory of automatic information processing
in reading. Cognitive Psychology, 6, 293-323.
Lesgold, A. M., & Perfetti, C. A. (1981). Interactive processes in reading: Where do we stand?
In A. Lesgold & C. Perfetti (Eds.), Interactive processes in reading (pp. 387-405).
Hillsdale, NJ: Erlbaum.
MacGinitie, W. H., & MacGinitie, R. K. (1978). Gates-MacGinitie Reading Tests: Canadian
Edition. Toronto: Nelson Publishing.
Mann, V. A., Cowin, E., & Schoenheimer, J. (1989). Phonological processing, language
comprehension, and reading ability. Journal of Learning Disabilities, 22, 76-89.

Applied Psycholinguistics 17:1 84
Young et al.: Effects of modeling and repeated reading
McMahon, M. L. (1983). Development of reading-while-listening skills in the primary grades.
Reading Research Quarterly, 19, 38-52.
O'Shea, L. J., Sindelar, P. T., & O'Shea, D. J. (1985). The effects of repeated readings
and attentional cues on the reading fluency and comprehension. Journal of Reading
Behavior, 17. 129-142.
(1987). The effects of repeated readings and attentional cues on the reading fluency and
comprehension of learning disabled readers. Learning Disabilities
Research,
2, 103-109.
Perfetti, C. A. (1985). Reading ability. New York: Oxford University Press.
Perfetti, C. A., & Lesgold, A. M. (1979). Coding comprehension in skilled reading and
implications for reading instruction. In L. Resnick & P. Weaver (Eds.), Theory and
practice of early reading (Vol. 1, pp. 57-84). Hillsdalc, NJ: Erlbaum.
Rasinski, T. V. (1985). The study of factors involved in reader-text interactions that contrib-
ute to fluency in reading. Unpublished doctoral dissertation. Ohio State University,
Columbus.
(1990). Effects of repeated reading and listening-while-reading on reading fluency. Journal
of Educational Research, 83. 147-150.
Rashotte, C. A., & Torgesen, J. K. (1985). Repeated reading and reading fluency in learning
disabled children. Reading Research Quarterly, 20, 180-189.
Reitsma, P. (1988). Reading practice for beginners: Effects of guided reading, reading-while-
listening, and independent reading with computer-based speech feedback. Reading Re-
search Quarterly, 23, 219-234.
Samuels, S. J. (1979). The method of repeated readings. Reading Teacher, 32, 403-408.
Schreiber, P. A. (1980). On the acquisition of reading fluency. Journal of Reading Behavior,
12, 177-186.
(1986).
On the role of prosody in children's syntactic processing (Program Report 86-11).
Madison: Wisconsin Center for Education Research.
(1987). Prosody and structure in children's syntactic processing. In R. Horowitz & J. Sam-
uels (Eds.), Comprehending oral and written language (pp. 243-270). San Diego, CA:
Academic.
Spring, C, Blunden, D., & Gatheral, M. (1981). Effect on reading comprehension of training
to automaticity in word-reading. Perceptual and Motor Skills, 53, 779-786.
Stanovich, K. E. (1986). Matthew effects in reading: Some consequences of individual differ-
ences in the acquisition of literacy. Reading Research Quarterly, 21, 360-407.
Swalm, J. E. (1972). A comparison of oral reading, silent reading, and listening comprehen-
sion. Education, 92, 111-115.
Taylor, N. E., Wade, M. R., & Yekovich, F. R. (1985). The effects of text manipulation and
multiple reading strategies on the reading performance of good and poor readers.
Reading Research Quarterly, 20, 566-574.
Turner, A.,
&
Greene, E. (1978). The construction of a propositional text base. JSAS Catalog
of Selected Documents in Psychology, 8, 1-81.
van Bon, W. H., Boksebeld, L. M., Freide, T. A., & van den Hurk, A. J. (1991). Journal of
Learning Disabilities, 24, 471-476.
Woodcock, R. (1987). Woodcock Reading Mastery Test-Revised. Circle Pines, MN: Ameri-
can Guidance Service.
Young, A. R., & Bowers, P. G. (1995). Individual difference and text difficulty determinants
of reading fluency and expressiveness. Journal of Experimental Child Psychology, 60,
428-454.