WATERFORD ASSESSMENT OF CORE SKILLS
A Computerized Adaptive Reading Test for Pre-K through 2
nd
Grade
Haya Shamir, Erin Phinney Johnson and Kimberly Brown
WaterfordResearch Institute, 55 West 900 South, Salt Lake City, UT, U.S.A.
Keywords: Computerized Adaptive Test, Reading Assessment.
Abstract: The Waterford Assessment of Core Skills (WACS) is a new computerized adaptive test of early literacy for
students in Kindergarten through 2
nd
grade. WACS includes assessments in letter recognition, letter sound
and initial sound recognition, blending, segmenting, reading real and non-words, reading comprehension,
listening comprehension, and vocabulary. A CAT for this age group will be highly beneficial by allowing
whole classes to be tested together without additional personnel, by assessing a large number of content
areas in reduced time and with fewer questions than a standard paper and pencil test, by producing
immediate and accurate score reports, and by engaging students with animations during the test. Reliability
and validity analyses indicate that the test is internally coherent and that the subtests correlate well with
other reading tests used with this age group, including DIBELS, IRI, ITBS and TPRI.
1 INTRODUCTION
For the last 33 years the Waterford Research
Institute has strived to develop high-quality
educational models and programs to enable all
children to receive the finest education possible.
Over time it has become evident that the ability to
easily assess student skills in the youngest grade
school group, when students are most responsive to
intervention, is lacking. Many current assessments
available for this age group, Kindergarten through
2
nd
grade, require one-on-one administration (e.g.,
DIBELS; Good & Kaminski, 2003), which may
result in a great deal of error variance due to
differing administration techniques by individual
testers or differences in scoring ambiguous answers.
In addition, tests that do not require one-on-one
administration are limited in the scope of what they
can cover and they risk introducing too much
variance from fidgety 6-year-olds (e.g., ITBS;
Hoover, et al., 2003). A computerized test, such as
the Waterford Assessment of Core Skills could
provide consistent and efficient test administration
by removing the need for a human test administrator
and providing an engaging testing environment.
2 METHODS
The Waterford Assessments of Core Skills (WACS)
is a web-based adaptive literacy test for pre-
kindergarten to second grade students. The new test,
soon to be available to schools and home users,
assesses early literacy skills including letter
recognition, letter sound and initial sound
recognition, blending, segmenting, the reading of
real words, non-words, and sight words, and
comprehension of paragraph-level text. WACS also
assesses early language skills including vocabulary
and listening comprehension. As an adaptive test,
WACS can assess a large number of content areas in
reduced time and with fewer questions than a
standard paper and pencil test. In addition,
computerized adaptive tests (CATs) may reduce
frustration for lower performing students and
boredom for higher performing students.
2.1 Design
The award winning product design team at
Waterford (Software and Information Industry
Association, 2008) have created an engaging test
that students actually enjoy taking. Throughout the
test students are guided by a groundhog named
Wyatt (see Figure 1) who is asking for their help on
a number of tasks they have to do together. At the
25
Shamir H., Johnson E. and Brown K. (2009).
WATERFORD ASSESSMENT OF CORE SKILLS - A Computerized Adaptive Reading Test for Pre-K through 2nd Grade.
In Proceedings of the First International Conference on Computer Supported Education, pages 24-30
Copyright
c
SciTePress
end of each section and at the end of the assessment
the students receive a non-judgmental reward screen
that serves as a short mental break of dancing
characters, with a new character appearing after each
skill has been completed. When students have fully
completed WACS, Wyatt presents them with a
deputy badge as a reward for finishing the
assessment.
WACS may be used by home users as well as
school users, allowing home-schooling parents
access to assessment tools similar to those used in
the public or private schools. Home users will be
able to complete the test by streaming the required
media, while schools will be expected to download
the related media to the computers in the classrooms
or computer labs.
Figure 1: Wyatt, demonstrating the Sight Words
Assessment.
2.1.1 Item Design
All test items are presented visually and aurally but
do not require the child to speak the answer. For
Letter Sound, Real Words, Sight Words, and
Nonwords, the letter/word appears on the screen
with three speakers underneath (see Figure 1). Each
speaker says a different letter/word name. The
student must click on the correct speaker to match
the word or letter that is on the screen.
Vocabulary differs slightly from this
arrangement with a sentence missing one word
written at the top of the screen. The sentence is read
to the child and the child must pick a word from the
speakers that best completes the sentence.
In the case of Letter Recognition, Initial Sound
and Blending there is a single speaker or picture at
the top of the screen and three pictures or letters at
the bottom of the screen (see Figure 2 for example).
The speaker/picture emits a sound or a series of
sounds and the student must select the picture below
that begins with that same sound, matches the series
of sounds, or select the letter that matches the letter
name from the speaker.
Figure 2: Example of the Initial Sound Assessment. The
answer is selected by clicking on the chosen letter with the
cursor, the green arrow.
For Reading and Listening Comprehension the
student is given a passage to read or listen to. When
finished, the child is presented with a question about
the passage followed by three possible answers. In
Listening Comprehension the questions and answers
are presented aurally. Each reading or listening
passage includes four questions of varying difficulty.
Each child receives three passages depending on
skill level.
The final task, Segmenting, differs from all of
the other tasks. Here, a picture is presented to the
student and he/she must move a series of blocks
representing the sounds into the correct order for the
word associated with the picture.
For all skills, the computer introduces the
question including the correct answer and the
distracters. The student can use the mouse to roll
over the question or the answer options to hear the
instructions again.
2.1.2 Sequence Design
Importantly, all children do not receive all
assessments. WACS includes eleven different
assessments, a subset of which is given to students
depending on their grade level and performance.
Limiting the number of assessments completed for
each child continues to reduce time required to test
the student and allows for a more pinpointed report.
It is assumed that students who are advanced in
reading do not need to be tested on pre-reading skills
such as letter recognition. However, students from
advanced grades may receive basic skills if they fail
CSEDU 2009 - International Conference on Computer Supported Education
26
to complete advanced skills at their grade level. On
the other hand, advanced students at lower grades
may receive more advanced skills if they prove to be
competent at the more basic skills.
2.1.3 Report Design
One of the greatest benefits of computerized testing
for students is the ability to receive scores
immediately after testing has completed. WACS has
been designed to provide reports about individual
test takers, as well as class, school, and district level
reporting. The reports indicate the child’s grade
level for each of the completed assessments as well
as detailed information about what was actually
being assessed and ways that any problem areas
could be addressed in the home or in the classroom.
The past three test results for the students are also
generated on the report, allowing parents and
teachers to compare changes over time in relevant
assessments.
2.2 IRT Analysis
In September, 2007, 8,800 students in Utah, Idaho,
Nevada, California, New York, Texas, North
Carolina, and Florida completed the first round of
testing with WACS. This first group was given a
random sample of questions from each assessment,
all questions representing varied expected difficulty
levels. The sample of students from twenty six
schools was representative of US socio-economic
status, ethnicity, geographic location, and type of
school, based on information obtained from the US
2002 census.
Based on the responses, difficulty values for
these 2,680 items were calibrated using the Rasch
model analysis for item response theory. Results
revealed 131 items with an outfit mean square
greater than 1.7, indicating high error variance in the
item. These items were excluded from the test. An
additional 131 items with outfit mean square less
than .5 were excluded, since items with outfit mean
square smaller than .5 are considered less productive
to the measure. Subsequent differential item
functioning (DIF) analysis revealed 21 items that
had a gender bias. These items were removed for
content review. Item difficulty was then calculated
for the remaining items.
Utilizing IRT analysis on test items, the adaptive
nature of the test allows a student’s response to
determine the next set of items. For example, if a
student fails to answer a question correctly within a
skill area, the next question he receives will be less
difficult. If the student answers that second question
correctly, the next question is harder, but not as hard
as the previously missed question. In this way, a
computerized adaptive test identifies the student’s
skill level in a particular area. Because WACS can
test up to eleven different areas, detailed information
about the student’s abilities are subsequently
available to teachers and parents.
2.3 Validity
Validity is the argument that a specific test score
interpretation or use is valid. In other words, a test is
valid when it does what it is supposed to do. There
are three major categories of validity: those
associated with content, criterion, and construct.
2.3.1 Content Validity
In order to establish content validity, this paper
discusses the reasons for the test design and content
as well as the association between the given test and
state standards or curricula. First, content experts
investigated the most important skills for pre-
kindergarten through 2
nd
grade students and
established guidelines for writing items based on
published research. The areas covered included
acquisition of letter names and sounds (Adams,
1994; Evans, 2005), early phoneme awareness
(Wilson, 1996), sight word reading (Carroll, 1971;
Wilson, 1996), real and non-word reading (Wilson,
1996; Ganske, 2000), vocabulary (Stemach &
Williams, 1988; Beck, McKeown, and Kucan,
2002), and reading and listening comprehension
(Snow, Burns, & Griffin, 1998). In addition, all
comprehension passages and questions were written
by professional writers, reviewed by content experts,
and edited by writing experts. Reading
comprehension passages were Lexile certified for
their grade levels.
Upon completion, items from all of the subtests
were reviewed by additional content experts and sent
to Marilyn Jager Adams, an external content expert,
for review. After IRT testing, analysis was
conducted to insure that item difficulty, as
determined statistically by IRT analysis, correlated
with the item difficulty as determined by the content
experts.
In addition to creating items based on researched
concepts, a valid reading test should also cover
standards accepted by the states for reading and
language development. Thus, state standards were
examined and correlated with WACS skills and
items. With the exception of Iowa (which did not list
WATERFORD ASSESSMENT OF CORE SKILLS - A Computerized Adaptive Reading Test for Pre-K through 2nd Grade
27
standards below grade three), a minimum of three,
and a maximum of eleven, WACS assessed skills
were also listed as state education standards for
PreKindergarten through grade two.
2.3.2 Criterion-related Validity
The effectiveness of a test in predicting performance
on a related task can be measured by assessing
performance on two tests at the same point
(concurrent validity) or at two different time points
(predictive validity). To assess concurrent validity
WACS was administered to students nationwide in
September and October. Student performance was
then compared to performance on the Dynamic
Indicators of Basic Early Literacy Skills (DIBELS),
the Idaho Reading Inventory (IRI), the Iowa Test of
Basic Skills (ITBS), and the Texas Primary Reading
Inventory (TPRI). Additional test data for the
Stanford Achievement Test (SAT 10) and ITBS will
be collected in April, 2009.
Data for predictive validity will also be collected
in the spring. WACS will be administered to the
same students from the fall testing. Those students
are also completing a spring round of testing for
DIBELS, IRI, or TPRI, making it possible to
examine predictive validity as well as a second
assessment of concurrent validity.
All five tests used to measure WACS validity
examine early reading skills and some include
subtests similar in name and concept to those given
in WACS, providing a stable comparison.
2.3.3 Construct Validity
Construct validity is typically measured with factor
analysis or principle component analysis. Because
the data was acquired with a computer adaptive test,
the large amount of missing data makes a typical
factor analysis less useful. Instead, a Rasch Factor
Analysis, completed in WINSTEP software,
performs a principle component analysis in order to
verify that our data is unidimensional. An additional
modified factor analysis is run with SPSS for
confirmation.
2.4 Reliability
Reliability refers to the consistency of a measure;
tests that have adequate reliability will yield more or
less the same scores across periods of time and
across different examiners. Because WACS is
administered on the computer there is no error
generated from different examiners. However, error
may still be introduced into the resulting final scores
through lack of attention to the task at hand, faulty
headphones, and disinterest. Because of these
concerns, it is important to examine test-retest
correlations with a small gap between testing dates
as well as the internal consistency of the test.
Computerized adaptive tests differ on measures
of test-retest reliability since an individual does not
see the exact same test at each time point. The
resulting correlation coefficient is regarded as a
conservative estimate since content sampling adds
an extra degree of error beyond individual
performance. CATs also differ on measures of
internal consistency. Traditional methods, split-half
reliability and Cronbach’s Alpha, are statistically
inaccurate when applied to a CAT tailored to
achievement. Instead, the marginal reliability
coefficient provides a better measure of internal
consistency by combining measurement error
estimated at multiple points on the scale. The
resulting coefficient is almost identical to
Cronbach’s alpha.
3 RESULTS
3.1 Concurrent Validity
Concurrent validity analyses were performed with
DIBELS, IRI, ITBS, and TPRI. The Kindergarten
WACS combination of tasks includes Blending,
Initial Sound, Letter Recognition, Letter Sound, and
Vocabulary. The first and second grade
combinations include Real Words, Nonwords, Sight
Words, Reading Comprehension, and Vocabulary.
Overall, correlations between relevant WACS
assessments and the associated paper and pencil test
are highly significant (Table 2), even with the
currently low number of participants taking the
ITBS (see Table 1).
Table 1:Number of students completing comparison test.
Assessment
N
Kindergarten
N
First
N
Second
DIBELS 206 142 137
IRI 120 61 126
ITBS 66 69
TPRI 170 155 190
DIBELS Beginning Kindergarten assessment,
consisting of Letter Naming Fluency and Initial
Sound Fluency, significantly correlates with WACS
Kindergarten Skills (
r = .74, p < .001). Similarly,
DIBELS Beginning First Grade assessment,
consisting of Letter Naming Fluency, Phoneme
CSEDU 2009 - International Conference on Computer Supported Education
28
Segmentation Fluency, and Nonword Fluency,
significantly correlates with WACS 1
st
grade Skills
(r = .72, p < .001) and DIBELS Second grade
assessment, consisting of Nonword Fluency and
Oral Reading Fluency, correlates with WACS 2
nd
Grade Skills (r = .61, p < .001).
Patterns or correlations for the IRI are similar to
those seen with DIBELS. The IRI includes only one
test for Kindergartners, the Letter Naming Fluency
test, and this task correlates significantly with
WACS Kindergarten Skills (r = .57, p < .001). First
grade IRI tasks, Letter Naming Fluency and
Nonword Fluency, also correlate highly with WACS
1
st
grade Skills (r = .74, p < .001). Finally, second
graders taking the IRI receive only the RCMB, a
reading fluency task. This IRI reading task also
significantly correlates with WACS 2
nd
grade Skills
(r = .58, p < .001).
The ITBS includes a number of areas of
assessments. For our purposes, comparisons are only
made to the Reading subtest for 1
st
and 2
nd
graders.
ITBS Reading significantly correlates with WACS
1
st
grade Skills (r = .7, p < .001) as well as with
WACS 2
nd
grade Skills (r = .41, p < .001).
The TPRI is designed for students to receive
additional assessments based on previous
performance. Only Letter Sound, Blending, Letter
Name, and Comprehension are given to all
Kindergartners. This combination significantly
correlates with the WACS Kindergarten Skills (r =
.52, p < .001). The TPRI combination given to all
first graders includes Letter Sound, Word Reading,
Word Per Minute Rate, and Comprehension
Questions. This combination also significantly
correlates with the WACS 1
st
Grade Skills (r = .64, p
< .001). Finally, the TPRI combination given to all
second graders, including Word Reading, Words
Per Minute Rate, and Comprehension, also
significantly correlates with WACS 2
nd
Grade Skills
(r = .5, p < .001).
Due to the way the test sequencer works, very few
students were given the Segmenting assessment this
fall. As a result, Segmenting could not be added to
any of the combinations for the grades. However, in
order to more thoroughly understand how this task
correlates with other reading-related tasks,
Segmenting was correlated with two relevant tasks
for which the n was over 30. Segmenting correlated
significantly with both DIBELS Nonword Fluency (r
= .42, p < .05) and IRI Reading (r = .37, p < .05).
Table 2: Relevant correlations between WACS and school-administered assessments.
WACS Assessments
WACS 1
st
Grade Skills WACS 2
nd
Grade Skills WACS Kindergarten Skills
r = .74, p < .001 DIBELS, Beginning
Kindergarten
DIBELS, Beginning 1
st
Grade
r = .72, p < .001
DIBELS, Beginning 2
nd
Grade
r = .61, p < .001
r = .57, p < .001 IRI, Kindergarten
IRI, 1
st
Grade r = .74, p < .001
IRI, 2
nd
Grade r = .58, p < .001
r = .52, p < .001 TPRI, Kindergarten
TPRI, 1
st
Grade r = .64, p < .001
TPRI, 2
nd
Grade r = .50, p < .001
ITBS 1
st
Grade r = .70, p < .001
ITBS 2
nd
Grade r = .41, p < .001
WATERFORD ASSESSMENT OF CORE SKILLS - A Computerized Adaptive Reading Test for Pre-K through 2nd Grade
29
3.2 Construct Validity
In general, when over 60% of the variance is
explained by a single factor, a test is considered to
have only one underlying factor. For WACS, 63.5%
of the variance is explained by a single factor.
Unexplained variance within the 1st factor is 4.5 %
(unexplained variance smaller than 5% confirms that
there is a single factor). An additional modified
factor analysis run with SPSS produces similar
results, with the first factor explaining 60.5% of the
variance and the next highest factor only explaining
9.7% of the variance. All assessments load strongly
on the first factor (all weights above .63) and only
Letter Recognition and Letter Sound have weights
above .4 on the second factor. In addition, the scree
plot indicates a dramatic drop from the first (eigen
value of 6.7) to the second factor (eigen value of 1).
Finally, another test of the internal coherence of
WACS overall is to examine correlations between
subtests. Resulting correlations indicate significant
relationships among all of the WACS subtests,
ranging from r = .38 (between Letter Recognition
and Listening Comprehension) to r = .74 (between
Letter Sound and Initial Sound), supporting the
conclusion that all subtests can be grouped together
as a unidimensional test.
3.3 Reliability
Test-retest correlations will be completed in April,
2009, when students take their spring WACS test.
The preliminary reliability correlation for WACS
Kindergarten Skills, with a sample size of 127, was
significant (r = .52, p < .001) as was the reliability
correlation for WACS 1
st
Grade Skills, with a
sample size of 85 (r = .73, p < .001).
Internal reliability has already been measured
with the marginal reliability coefficient, examining
internal test consistency. Reliability for WACS, is
very strong (r = .93).
4 CONCLUSIONS
Currently, very few standardized assessments are
capable of being used in for pre-K through 2
nd
grade
educational group. Even tests that can be used with
these young children often aren’t used, likely due to
difficulties in keeping young children engaged.
WACS has been designed specifically for young
children, and by presenting the testing information
on the computer, the children are able to stay
engaged with the animated characters. Current No
Child Left Behind standards require testing
beginning in third grade. However, research
demonstrates that early detection and intervention
are essential for academic success. Identifying
struggling students early increases these student’s
chances of being successful readers and meeting the
NCLB requirements. With an assessment that is easy
to administer, engaging for the students and provides
accurate immediate results, more students are likely
to be reading at or above grade level in the future.
With validation on the Waterford Assessment of
Core Skills completed this coming spring, WACS
will become an important part of grade school
education.
REFERENCES
Adams, Marilyn, Jager. (1994). Beginning to Read:
Thinking and Learning about Print. Cambridge,
Massachusetts: The MIT Press.
Beck, Isabel, L., McKeown, Margaret, G., Kucan, Linda.
(2002). Bringing words to life: robust vocabulary
instruction. New York, New York: The Guilford Press.
Carroll, J. B., Davies, P., and Richman, B. (1971). The
American Heritage Word Frequency Book. New York:
Houghton Mifflin.
Evans, Mary Ann. (2005 June). Phonological awareness
and the acquisition of alphabetic knowledge. Paper
presented at the twelfth annual meeting of Society for
the Scientific Study of Reading, Toronto, Canada.
Ganske, Kathy. (2000). Word journeys: assessment-guided
phonics, spelling, and vocabulary instruction. New
York, New York: The Guilford Press.
Good, R. H., & Kaminski, R. A. (2003). Dynamic
indicators of basic early literacy skills, 6th edition.
Eugene: Institute for the Development of Educational
Achievement, University of Oregon
Hoover, H. D., Dunbar, S. B., Frisbie, D. A., Oberley, K.
R., Ordman, V. L., Naylor, R. J., Lewis, J. C., Quails,
A. L., Mengeling, M. A., & Shannon, G. P. (2003).
The Iowa Tests, guide to research and development,
Forms A and B. Itasca, IL: Riverside Publishing.
Snow, C. E., Burns, S., & Griffin, P. (1998). Preventing
Reading Difficulties in Young Children. Washington,
D.C., National Academy Press.
Software and Information Industry Association. (2008).
CODiE Awards, Best Course/ Classroom Management
Solution and Best Science Instructional Solution.
http://www.siia.net/codies/2008/winners
Stemach, J. & Williams, W. B. (1988). Word Express: The
First 2,500 Words of Spoken English. High Noon
Books.
Wilson, Barbara, A. (2005, November). Decoding.
Symposium conducted at the meeting of the
International Dyslexia Association, Denver, Colorado.
Wilson, Barbara, A. (1996). Wilson Reading System
Instructor Manual. Boston, Massachusetts: Wilson
Language Training Corp.
CSEDU 2009 - International Conference on Computer Supported Education
30