Download as pdf or txt
Download as pdf or txt
You are on page 1of 169

MATH COMPUTATION DIFFICULTIES IN GRADE 7 AND 8 STUDENTS

Cynthia Ann McCall

A thesis submitted in conformity with the requirements


for the degree of Doctor of Philosophy
Department of Human Development and Applied Psychology
Ontario Institute for Studies in Education of the
University of Toronto

8 Copyright by Cynthia AM McCall ( 1999)


I*I oNational
f Canada
Library Bibliothèque nationale
du Canada
Acquisitions and Acquisitions et
Bibliographie Services services bibliographiques
395 Wellington Street 395. nie WeIIington
OttawaON K l A O N 4 Ottawa ON K1A ON4
Canada Canada
Your Votre nikreoce

Our füe Norre reference

The author has granted a non- L'auteur a accordé une licence non
exclusive licence allowing the exclusive permettant à la
National Library of Canada to Bibliothèque nationale du Canada de
reproduce, loan, distribute or sel1 reproduire, prêter, distribuer ou
copies of this thesis in microforni, vendre des copies de cette thèse sous
paper or elechonic formats. la fome de microfiche/film, de
reproduction sur papier ou sur format
électronique.

The author retains ownership of the L'auteur conseive la propriété du


copyright in this îhesis. Neither the droit d'auteur qui protège cette thèse.
thesis nor substantial extracts fkorn it Ni la thèse ni des extraits substantiels
may be printed or othenvise de celle-ci ne doivent être imprimés
reproduced without the author's ou autrement reproduits sans son
permission. autorisation.
Math Computation Difficulties in Grade 7 and 8 Students

Cynthia Ann McCall, Ph.D., 1999


Department of Human Development and Applied Psychology
Ontario Institute for Studies in Education of the University of Toronto

ABSTRACT

Little research has been done on the underlying factors contributing to math

computation difficulties. This study investigated some of the cognitive abilities and basic

mathematics skilfs that are believed to be related to success in math computations. From

a pool of 187 grade 7 and 8 students, thirty students were identified as having math

disabilities (MD) based on the WRAT3 Anthmetic standard scores. Thirty peers who

achieved WRAT3 Arithmetic, Reading and Spelling standard scores within the Average

range were selected as members of the normal control (NC)group. Comparisons were

made between the two groups on variables thought to be related to math computation:

processing speed, memory, basic numeration concepts and number facts.

The results revealed that the NC Group scored significantly higher than the MD

Group on the measures of basic numeration concepts such as place value, estimation and

number patterns. Many of the procedure errors made by the MD Group could be related

to these concepts. The NC Group also outperfomed the MD Group on al1 of the

processing speed measures and on a sequential memory task. The processing speed

measures, dong with the numeration concepts and nurnber fact autornaticity level,

predicted a significant arnount of the variance in the WRAT3 Arithmetic scores.

Surprisingly, no group differences were noted on the working memory tasks, nor were

verbal short-term memory differences found.


As expected, the NC Group knew significantly more number facts autornatically

than the MD Group in al1 four operations. There was however, growth in the MD Group

across grades. The pattern of automaticity was similar across operations and groups, with

ties being easier in most cases. Both groups used the same type of back-up strategies for

non-automatic facts.

The possibility of heterogeneity within the MD Group was explored by dividing

this group of students into subtypes based on their academic profiles. The expected

pattern of visual-spatial deficits in the students with exclusively arithmetic disabilities

(Group A) was rare. Few differences were found on the cognitive and math measures

between students in Group A and ihose with deficits in arithmetic, reading and spelling

(Group ARS).
Firstly, 1 owe my greatest debt to Dr. Dale Willows, whose encouragement,
guidance, availability day or night and willingness to go above and beyond the cal1 of
duty enabled me to complete this dissertation. Her knowledge and high standards
coupled with an approachable manner are an impressive combination. These qualities
allowed me to view the dissertation process positively and with a sense of satisfaction.
1 am most emteful to Dr. Tom Humphries, whose clarity of thought and ability to
distill large volumes of information into a succinct f o m was invaluable. Dr. Humphries'
generosity of time and expertise from the very beginning of my post graduate work
played an enormous role in moving me almg the path towards my Ph.D.
Dr. Rosemary Tannock's positive attitude, thought provoking questions and
ability to link my study to other areas of research were a great asset and deeply
appreciated.
Thanks to rny committee for willingly taking on a subject area that was not the
main focus of their own research. They thoughtfully applied their own expertise to the
area of math and in so doing expanded my view on the subject. In spite of their own
pressures and tirnelines they gave unselfishly of their time and energy whenever it was
needed. I feel very privileged to have had this team of individuals to work with who
possess such generosity of spirit.
I am grateful to Dr. John Elkins, who despite living half way around the world
was willing to take on the roie of extemal examiner and provided insightful cornments on
my research. 1 am indebted to Dr. Glenn Regehr who initially guided me through the
statistical process and subsequently was a member of my committee.
I extend my gratitude to Dr. Howard Marcovitch who helped me see the forest
when 1 became too focused on the trees. 1 was indeed the beneficiary of his flexibility
and technical and moral support. My thanks to Dr. Ruthann Hicks for her help in steering
me through the necessary procedures in the initial stages of this study and her continued
interest throughout.
I would like to thank: Peter for his statistical hetp; my good friend Jane for her
editing skills; my aisle-mates at work, Paul and Anne, w ho willingly intempted their
work to listen to my ideas and contribute their thoughts; rny OISE buddies, Donna for
sharing what she learned from her own dissertation experience and Karen who made the
dissertation process more fun, as we went through it topther. To my fnends and family,
thanks for your support, the hibernation phase is now over.
Finally, 1 would like to thank the principals of the schools that were involved and
al1 of the grade 7 and 8 students whose willingness to participate in this study made the
research possible.
TABLE OF CONTENTS

CWPTER 1
INTRODUCïION ..............................................................................
Purpose..................................................................................
Review of Literature ...................................................................
Math Computation............................................................
Error Classification ...................................................
Computation and Math DisabiIities ................................
Basic Math Concepts.........................................................
Decomposing and Recomposing Numbers ........................
Place Value ............................................................
Estimation .............................................................
Basic Math Concepts and Math Disabilities ......................
Nurnber Facts ..................................................................
Number Fact Strategies........... ...................................
Number Fact Characteristics........................................
Number Fact Errors..................................................
. .Number Facts and Math Disabilities- .............................
Cogmtive Variables............................................................
Memory ................................................................
Processing Speed......................................................
Math Disability Subtypes.....................................................
Arithmetic Patterns by Subtype.....................................
Limitations to Previous Research ............................................
Overview of Previous Findings ...............................................
Rationaie for the Study........................................................

CHAETER 2
MErHOD .........................................................................................
Subjects ..................................................................................
Procedure................................................................................
Measures.................................................................................
Intellectual Potential Measure...............................................
Written Language Academic Measures ....................................
Math Measures .................................................................
Memory Measures.............................................................
Processing Speed Measures ..................................................

CHAPTER 3
RESULTS AND DISCUSSION............................................................... 46
Section 1.......................................................................................... 46
Dernographic Information.............................................................. 46
Cornparisons Between the NC and MD Gtoups.............. . ......... 46
Cornparisons between the NC. A and ARS Groups ....................... 48
Statistical Information.................................................................. 49
Normality and Homogeneity of Variance ...................................
Siatistical Analyses ............................................................
Section 2..........................................................................................
Cornparisons Between the NC and MD Groups .............................................
Analyses of Computation Skills......................................................
Computation Skills............................................................
Analyses of Computation Errors .............................................
Analyses of Basic Math Concepts ....................................................
Analyses of Number Fact Data ........................................................
Timed Number Fact Mastery................................... ..............
Number Fact Autornaticity...................................................
LeveI of Number Fact Automaticity for Each Question ..................
Back-Up Strategies.............................................................
Use of Each Type of Strategy .......................................
Counting Strategies in Addition and Subtraction.................
Reference Strategies Across Operations ...........................
harnples of Number Fact Back-Up Strategies and Errors ......
Number Fact Error Analyses .................................................
Analyses of Cognitive Variables ......................... ............................
Memory ................. . . ....................................................
Processing Speed.............................................. .................
Regession Analyses .....................................................................
Section 3...........................................................................................
Cornparison Between the NC, A and ARS Groups ..........................................
Compasison of Vocabulary and Block Design Discrepancy Patterns............
Resulîs from the NC, A and ARS Group Analyses .................................

GENIERAL DISCUSSION ..................................................................... Y7


Math Computation............... ... .
. . . ........................................... 97
Basic Math Concepts ................................................................... 99
Number Facts ............................................................................ IO1
Cognitive Variables .................................................................... 105
Predictors ................................................................................. 107
Math Disability Subtypes............................................................... lû8
A Potential Framework for Math Computations .................................... 109
Contributions............................................................................. 111
Limitations and Future Research................................................. 114

vii
LIST OF TABLES

Table 1 Stage 1 Groupings Based on WRAT3 Arithmetic and Spelling


Standard Scores................... , .......................................... 34
Table 2 Group Requirements............ .. ........................................... 36
Table 3 Test Battery..................................................................... 37
Table 4 Correlations for Non-Standardized and Standardized Math Measure... 41
Table 5 Characteristics of the Four Groups in Stage 2 of the Study ............... 47
Table 6 WRAT3 Anthmetic Item Analysis by Operation and Across Gro~ps ... 55
Table 7 The Percent of Students in Each Group Committing Each Type of
Computation Enor .......................... . . . . ............................ 59
Table 8 Comparison of NC and MD Groups on Measures of Basic Math
Concepts ............................................ ............ 61
Table 9 Comparison of the NC and MD Groups on the Numeration Subtest
Item Clusters ..................................................................... 62
Table 10 Comparison of the Groups on the Sequentiai and Quantitative
Reasoning Items .......................................................... 63
Table I l Percent of Each Strategy Used forAddition Number Facts ............... 70
Table 12 Percent of Each Strategy Used for Subtraction Number Facts ........... 70
Table 13 Percent of Each Strategy Used for Multiplication Number Facts ........ 70
Table 14 Percent of Each Strategy Used for Division Number Facts ....... ....... 70
Table 15 Number of Students Per Group Using Counting Strategies
in Subtraction......................... .,........................................ 76
Table 16 Number of students using Each Type of Reference Strategy for
Addition......................................................................... 78
Tabie 17 Number of students using Each Type of Reference Strategy for
Subtraction...................................................................... 78
Table 18 Number of students using Each Type of Reference Strategy for
MultipIication ................................................................... 78
Table 19 Number of students using Each Type of Reference Strategy for
Division ..............................,........................................ 78
Table 20 Cornparison of NC and MD Groups on Memory Tests ................... 82
Table 21 Comparison Between the NC and MD Groups on Items from the
Sequences Subtest...........................~...~.............................. 83
Table 22 Cornparison of MD and NC Students on Rocessing Speed Tests ....... 85
Table 23 Regression Analysis for WRAT' Anthmetic Without Math Measures. 89
Table 24 Correlation Matrix.............................................................. 90
Tabie 25 Regression Analysis for WRAT3 Arithmetic with d l Measures........ 91
Table 26 Number of Students in Each of the Vocabulary-Block Design Groups 93
LIST OF FIGURES

Figure 1 Schematic Representation of Computation Based on the Literature ... 31


Figure 2 A Comparison of WRAT3 Anthmetic Questions by Group. ........... 41
Figure 3 Examples of Computation Errors.. ......................................... 58
Figure 4 Cornparison of Number Fact Mastery Rate Between the NC and
MD Groups-.......................... . . . ..... - . .. .. .... ... ........... 66
Figure 5 Group by Grade Comparisons of Number Fact Mastery Rate.. ......... 67
Figure 6 Group by Grade Comparisons of Number Fact Automaticity ........... 71
Figure 7 Comparison of Number Fact Automatidy Between the NC and
MD Groups .............................. . ... . . . 72
Figure 8 Automatization of Number Facts by Number Fact Families.. .......... 74
Figure 9 Scores on the Cognitive and Academic Variables for the NC and
MD Groups-..-.-.... ..,..--...-...-.-....... .. . .. . . . . . ... . . 87
Figure IO Scores on the Cognitive and Academic Variables for the NC,A
and ARS Groups.. ................. ............ . .. ............ %
Figure 11 Areas of Potential Difficulty Within the Math Computation Schemata 111
LIST OF APPENDICES

Appendix A
Consent form......................................................................
Investigator Created Estimation Test ...........................................
Investigator Created Place Value Test .........................................
Investigator Created Stage 1 Number Fact Tests.............................
Investigator Created Stage 2 Number Fact Tests.............................
Investigator Created Rapid Number Naming Test ...........................

Appendix B
Number Facts ......................................................................
Level of Automaticity and Rankings for all Four Operations......
Spearman Correlations of Automaticity Rankings ..................
Number Fact Errors...............................................................
Number Fact Errors By Strategy................................................
Subtraction Number Fact k o r s................................................
Multiplication Number Fact Errors.............................................
Distance of Number Fact Response from the Correct Answer
in Each Operation.................................................................

Appendix C
Subtype Comparisons............................................................
Math Computation .......................................................
Basic Math Concepts ....................................................
Number Facts............................................................
..
ComphveMeasures ............. . ........... .., .....................
Memory .........................................................
Processing Speed ...............................................
INTRODUCTION

Pnrpo=
The acquisition of arithmetic computation skilis is important for the development
of more complex mathematical skills. However, it is this aspect of mathematics that
students with l e d n g disabilities often have the most difficulty consolidating (Akeman
et ai.. 19%; Badian, 1983; Reischner, Garnett & Shepherd, 1982). McLeod and Crump
(19'78) found that one tenth of eiementxy school children with learning disabilities had
severe deficits in math, but by the intemediate and secondary levels over 50% of
children with learning disabilities required remedial math support. In a survey of special
education teachers at the intermediate and secondary levels, McLeod and Armstrong
(1982) reported that 40.3% of the students were identifid as having arithmetic
disabilities with or without difficulties in other areas. Badian (1983) determined that
6.4% of elementary students in a small town had mathematics difficulties in contrast to
4.9% with reading difficulties. Weinstein (1980) suggested that approximately 6% of
students display a specifk arithmetic disability. Other reçearchers have found similar
prevalence rates (Feogens & McKinney, 1981; Geary, 1993; Kosc, 1974; Light &
Defnes, 1995; N o m & Zigmond. 1980). Despite the comparable rates of math and
reading difficulties and the high rate of comorbidity between the two m a s , studies on
math disorders are sparse compared to what is available on reading disabilities (Ginsburg,
199'7; Sutaria, 1985).
The purpose of this investigation was to identify grade 7 and 8 students with math
computation difficulties (MD Group) and average computation skills (NC Group) and
compare them on a number of dependent variables thought to be related to math
computation. Six main questions guided the cornparisons between these groups:

(i) What aspects of computation do grade 7 and 8 MD students have trouble with
compared to NC students and do they display ordinary or unusual emor
patterns?
Do MD students understand basic math concepts as weU as NC students?
Are the MD Group's number fact speed, automaticity, back up shategies and
errors sirnilar to those of the NC Group and are they consistent across
operations?
Are there differences between the NC and MD Groups in the cognitive areas of
processing speed and memory?
How much of the WRAT3 Anthmetic score can be accounted for by the
cognitive and math variables included in this study?
Are there subtypes within the MD Gmup that display different patterns of
sîremgîhs and weaknesses?

These questions guide the organization of this thesis, with the sequeace of information in
each chapter following the order of the questions.

Review of LiteratnrP
(i) Math Compntation
Kosc (1974) classified six subtypes of dyscalculia, one of which was operational
dyscalculia, referring to impairrnents in the abiIity to carry out arithmetic computations.
In order to complete a wmputation successfully, children must be able to associate a
symbol (+, -, x, +) wwith a particular operation. They also ne& to remernber the sequence
of steps required to solve an algorithm and retrieve answers to the basic number facts
included in the question. F i d l y , children must be able to write the questions on paper
by lining the numbers up correctly (Ashcraft, 1992;Bley & Thomton, 1989;Temple,
1991). Difficulties can anse in any of these steps, however, children with math
disabilities seem to have the most trouble with the procedural aspect of mmputation as
well as automatic retrieval of nurnber facts (Geary, 199û; Geary, Widaman, Little, &
Couimier, 1987; Svenson & Broquist, 1975).
A few studies have investigated the development of math cornputation abilities
through changes in grade-quivalent scores on standardized tests. Cawley and Miller
(19û9)determined that at 8 years of age children with learning disabilities were at a grade
1 level on the Woodcock-lohnson Psychoeducational Assessment Battery (WJPB)
Calculations subtest (Woodcock & Johnson, 1977) and by 17 years of age they were at a
grade 5.8 level, indicating a growth of less than five years over a 9 year penod. Most
germane to the present study was the finding that 12 and 13 year old students with
learning disabilities (ID) achieved mean grade equivalent scores of 2.6 and 3.6,
respectively. Ackeman and Dykrnan (1995) found that adolescents had grade equivalent
scores between the 3d and 6'hgrade on the Wide Range Achievement Test - Revised
(WRAT-R)(Jastak & Wilkinson, 1984)Arithmetic subtest.
Cawley, Pamar, Yan and Miller (1996) compared the performance of 9 to 14
year oid students with and wiîhout leaming disabiiities (LD) on their investigator created
computation test They found no significant diference between the two groups at nine
years of age. However, the gap between the two groups widened with each passing year
so that by age 14, the students with LD were performing similarly to the 10 year old NC
students- This finding suggests that it t w k the students with LD five years to make the
gains NC students made in one year. Whereas the NC Group showed steady progress
with age, the LD Group's progress tended to be inconsistent and included more plateaus.
Ackeman, Dykmm and Petes (1977) noted that adolescents originaliy identifïed
in elementary school because of reading difficulties had WRAT-R Arithmetic subtest
scores lower than their reading scores. Thus îhey suggested that students with LD f . 1
further behind in mathematics as they grow older. However, other factors that could have
contributeci to their decline, such as the initial math scores of these students, what
measures were used and how much math teaching occurred while the students were
receiving support for their reading difficulties. were not given.
Taken together, these findings suggest that students with MD or LD make slower
progress in the area of math computation than NC students, reaching a plateau around the
junior grades. In order to determine what aspects of computation are hindenng the
success of students with MD, e m r analyses are generally d e d out.

Error Classification
A computation error anaiysis permits the identification of where a miscue might
have occurred in a snident's algorithm (procedure). Brown and Burton (1978) suggested
that errors are not random, but an indication that the student has one or more errors in
their computation process. Ginsburg (1981) classified errors into two main categories,
slips and bugs. Slips were errors in e x d o n that occurred even though the student
knew the procedure. These are often termed careless errors and usually include
calculation and sign (doing the wrong operation) errors. Research into number fact errors
suggests that they may not be careless, but reflect errors in sûategy implementation
(Cumming & Elkins, 1994). Patterns also exist with sign errors, with multiplication
tending to be confused with addition or division, and division tending to be confused with
subtraction or multiplication (Campbell, 1997). Bugs are procedural errors, such as
subtracting the smaller number from the larger number regardes of position, that
indicate an incomplete understanding of an algorithm or incorrect application of a rule or
strategy.
A number of attempts have been made to classify arithrnetic erroa (Engelhart,
197; Greenstein & Strain, 1977;Roberts, 1968; Saito, 1992; Spiers, 1981). Although
each model uses different labels for errors, consistent terms have been used here in order
to facilitate cornparison between the mdels. Roberts (1%8) identified four main types
of errors in grade 3 students: sign, cdculation (e-g., number fact and multi-digit
computation errors), algorithm and unknown errors. The results of the Roberts' study
revealed that shidents of al1 ability levels made calculation errors and that algorithm
errors were the most common at al1 but the lowest ability levels. A negative correlation
existed between arithmetic ability level and sign and unknown errors.
Engelhardt (1977)expanded Robert's model to 8 categories using the erron of
grade 3 and 6 students on the Staaford Diagnostic Anthmetic Test (Beatty, Madden, &
Gardner, 1966). In addition to sign,calculation and aigorithm e m n , Engiehardt
included regrouping, inversion (e-g., reversal of digit, subtracting first number from
second), incomplete, identity (e.g., 5 x 1 = 1) and O errors. He found that calculation
errors were committed the most often by al1 ability levels. but the highly competent math
students made almost no algorithm emrs.
Greenspan and Strain (1977) snidied the errors of 12 to 17 year old students with
learning disabilities on the KeyMath Diagnostic Arithmetic Test (Comoily, Nachtman, &
Pritchett, 1976) and also identiFred calcuIation, algorithm and inversion errors, but
included two other categones: spatial (e-g., pour alignment of columns) and detail enon
(e.g., omission of dollar signs and decimals).
The most detailed classification system was developed by Spiers (1987) based on
a review of the literature on acalculia (math disabilities). It has categories similar to
those determined by Englehardf but ad& place holder (e.g., O as a place holder in the
second row of two-digit multiplication questions), digit (e-g., omission of digif
perseveration) and symbol errors.

Cornputaiion and Math Disabitities


Only a few studies have examined the math computation skills of NC and MD
students in order to determine where diffculties &se. Russell and Ginsburg (1984)
found that grade 4 MD students made more addition and subtraction computation errors
than grade 4 NC students but not as many as grade 3 NC students. Al1 three groups made
the most errors in subtraction. The MD Group made relatively more calcuiation erroe
than both NC Groups, which was consistent with their poor number fact skius in this
study. They noted that the qpes of errors made by the MD Group were similar to those
made by the NC Group.
Saito (1992) found that grade 3 and 5 MD students made more errors on the
WRAT-R Arithmetic subtest than NC students and that advanceci subtraction questions
seemed to be the stumbling block for grade 5 MD students compared to NC students. It
should be noted that only one error was identified for each question, with Saito deciding
which seemed to be the most salient. In general, multi-digit subtraction seems to be
harder for students to master than multi-digit addition, but in both cases, regrouping
appears to cause the most difficulties in elementary students (Fuson et al., 1997).
The differences between the LD and NC students (ranging from 7 to 14 years of
age) in the Cawley et d.'s (19%) study was particularly evident on questions invoiving
regrouping, zero as a digit, multiplication and division. They noted that many of the
incorrect answers in multiplication and division were due to addition and subtraction
mistakes, respectively. They also found that children with learning disabilities made more
procedurd enors overall, but that these errors were more likely to occur on the simpler
items (e.g., subtraction with regrouping). Calculation e m n seemed to increase with the
magnitude of the numbers used and the number of steps in a procedure. Miller and
Milam (1987)atso noted that the majority of multiplication errors students with Iearning
disabilities made were due to inadequate knowledge of multipücation facts and poor
addition skills. Ackerman and Dykman ( 1995) investigated the WRAT-R Arithmetic
protocots of adolescents with a combination of aithmetic and reading disabilities and
found that they had trouble with multiplication and division as well as fractions.
Fuson et al. (1997) descnbed the strategies used by students to solve rnulti-digit
addition and subtraction questions, based on their observations at four projects around the
world. They ooted that some students treated these problems as if they were adding
separate columns of single digits (concatenated single-digits) as opposed to viewing the
colums as ones, tens, hundreds, etc. (collected multiunit method). This
conceptualization, which is often used when questions are presented verticaliy, is error
prone because it does not provide enough of a frarnework to guide procedure selection.
An innovative study by Raiinen and Ahonen (1995) investigated the relationship
between computation enors on the WRAT-R and RMAT (Rasaen, 1993)and reading
speed and accuracy in grade 3 to 6 Ftnnish students with and without MD. They used a
modification of Spiers (1987) classification system to identify the emrs. They found that
in both groups, only the number fact errors (calculated for multiplication) were
sigaificantly correlated with reading speed and accuracy. Classification of the LD Group
according to reading speed and accuracy levels reveaied that the slower readers made
more fact-retrieval errors. The authors suggested this may indicate a cornmon underlying
factor for reading speed and number fact retneval.
Despite the recognition that math computation is an area of difficulty for students
with math disabilities, as of yet, there is no overall mode1 that explains the computation
process or where it can break down. The research has tended to be more descriptive in
nature. It appears that students with MD initially have trouble with regrouping in
addition and subtraction, then with multi-digit multiplication and division and later with
fractions. As a result, even by the end of high school, they do not seem to have
progressed past the junior grade level on standardized tests. Their errors seem to occur
on the procedural and cdculation aspects of cornputation and be similar to those found in
NC students. Sorne of their procedurai errors may reflect poor conceptuaiization of the
numbers in mukti-digit questions.

(ii) Basic Math Concepts


Basic math concepts, also referred to as number sense, include understanding
symbols, counting, decomposing and recombining numbers (e.g,, 53 = 5 tens t 3 ones =
4 tens + 13 ones). place value, the properties associated with each operation (e.g., the
commutative property of addition [a + b = b + a]). and estimation (Greeno, 1991). Many
elementary school teachers tend to view arithmetic as a rote learning activity, with the
focus on dnil and practice (Thompson, 1992). However, instruction that emphasizes rote
drill seems to separate the computation skills from the math concepts (Davis &
McKnight, 1980). The math curriculum in many countrîes has also emphasized the
mernorization of computation procedures, with Iittle focus on the base-ten place value
system (Bednarz & Janvier, 1982; Fuson, 1990;Kouba et al., 1988). It has been
hypothesized that procedural errors rnay result from students relying excIusively on
memorized procedures, so that with understanding of the base-ten place value system
these algorithmic errors would be eliminated (Fuson, 1992; Fuson et. al., 1997).
If the basic concepts are understood, then a frarnework for constmcting
reiationships between procedures, rules and number facts can be developed (Hiebert &
Carpenter, 1992). Having an overall frarnework also provides more hooks on which to
hang new information thus making it easier to recall. The multiple connections in an
overall structure also ailow particular pieces to be used flexibly either to create new
strategies, or to modify old ones (Carpenter & Moser, 1984; Fuson & Briars, 1990;
Hiebert & Lefevre, 1986; Hiebert & Wearne, 1992). Knowledge that cornes frorn
understanding is easier to apply to unfamiliar questions than memorized procedural
knowledge (Hiebert & Wearne, 1%; Carpenter et al., 1997).
To date, no study has investigated the developmental stages of basic math
concepts beyond the basic counting principles used by young children (e.g., one-to-one
correspondence, stable order [number words always follow the same order: one, two,. ..],
cardinatity (the last number counted represents the totaI number of counted abjects],
abstraction [anything can be counted] and order irrelevance [objects can be counted in
any order] (Geary, 1994). Therefore, as with the previous section on computation, much
of the information presented here is descriptive in nature.

Decornposin~and Recom~osingNurnbers
The decomposing and recornposing of numbers has been related to Piaget's
conservation concepts. Both are built on an understanding of part-whole concepts and in
both cases the original value is transformed but the total arnount rernains the same (e.g.,
13 is transformed into 6 + 7;a tail, thin glass of water is poured into a short, fat glass)
(Derr,19û5). Although the results have been mixed, some studies indicate that students
with math disabilities lag behind their peers in the development of Piaget's conservation
concepts (Andersson et al., 1980; Derr, 1985; Saxe & Shaheen, 1981). This delay may
hinder the development of strategies for number fact acquisition. deBettencourt et al.
(1993) found that only some grade 3 and 5 students with leaming disabilities could give
complete explanations of derived fact strategies for addition (e-g., 5 + 6 = 5 t 5 + 1 = 11)
and even fewer for subtraction, compared to the grade 3 NC Group. The authors
suggested that the LD Group displayed a delay in their understanding of derived fact
strategies and appeared still to rely on counting strategies. A poor understanding of
decomposing and recombining may also affect the mastery of regrouping, as in order to
regroup, students need to understand how to decornpose numbers (e.g, tuming one ten
into ten ones) (Fuson et al., 1997). Computations that require regrouping also make the
most demands on base-ten place value knowledge (Hiebert & Wearne, 19%).

Place Value
Base-ten place value concepts have been related to rnany aspects of math
computation (Resnick, 1983). In particular, a poor understanding of place value has been
suggested as a factor in regrouping errors (Ginsburg, 1989; Russell & Ginsburg, 1984).
Hiebert and Wearne (19%) camed out a longitudinal study of students from grade 1
through to grade 4. A cornparison of students who understood place value concepts with
those who did not found that after instruction both groups selected appropriate procedures
for addition with repuping, but the understanders were better able to explain what they
were doing and could generalize the strategy to similar questions (e-g., subtraction with
regrouping). There was also some suggestion that nonunderstanders forgot the procedure
somewhat when it was not practiced regularly. The nonunderstanders were also
dependant on instruction for the development of procedures because they could not
generate them on their own as the understanders could.
Carpenter et al. ( 199'7) carrïed out a three year longitudinal study of children's
invented strategies for solving addition and subtraction questions and their relationship to
the developrnent of basic math concepts. They found that children who used invented
strategies developed knowledge of base-ten concepts earlier tban those who memorized
procedures and the former made fewer aigorithmic errors.
In a study by Mura and Okamato (1989) comparing U.S. and Japanese grade 1
students, the understanding of place vdue was correlated positively with the use of
decomposition strategies with ten (e.g. 7 + 6 = 7 + 3 [= 101 t 3 = 13) and negatively with
the use of counting strategies (7 + 6 = 7,8.9,. .-13)for both groups. Place value also
correlated positively with math achievement in both countries.

Estimation
A less studied basic math concept is estimation. When presented with math
computations, good estimators can use a variety of strategies, but poor estimators are
Limited to the algorithms they learned for paper and pencil computations (Rubinstein,
1985: Threadgill-Sower. 1984). Good estirnators round to numbers that make the mental
computations easier (Reys et al., 1982), thus they can overcome the high memory
demands produced by pencil and paper approaches. Markovits and Sowder (1994) found
that limited instruction Led to improvements in estimation abilities and an understanding
of the effects of estimation on answers, in twelve grade 7 boys.

Basic Math Concepts and Math Disabilities


Kouba et al. (1988) showed that U.S. children have difficulty with place value and
that many of the computation errors they make appear to be due to a lack of
understanding of this basic math concept. Cawley et al. (19%) also noted that many of
the algorithmic errors made by the 7 to 14 year old students with learning disabilities in
their study seemed to be linked to difficulties with place value.
Russell and Ginsburg (1984) compared grade 4 students with math difficulties to
nomally achieving grade 3 and 4 students on a number of basic math concepts. The MD
Group did significantly worse than theK peers. but similarly to the younger stuclents on a
counting task in which they had to sum piles of money consisting of ten. hundred and
thousand dollar bills. Almost half of the MD Group could not do this at al1 and for al1
students most of the errors were related to place value. The same pattern held when
students were asked to detemine how many of a smaller number were in a larger number
(e.g., how many 10's in 100). This task seems to have an aspect of estimation to it.
Differences were not found on the other place value tasks (e.g., determining the number
of dots [arranged in =pups of ten] on a card, identifying which of two numbers was
larger [e.g 499!399 versus 51111l), representing each digit in a number with chips). As
the place value difficulties only appeared to occur with iarger numbers, Russeil and
Ginsburg concluded that poor computation abilities did not stem from an inadequate
grasp of place value, but further research was suggested.
Geary, Bow-Thomas and Yao (1992) found that grade 1 students with math
disabilities had trouble detecting counting errors and had yet to determine which counting
features were essential versus unessential (e.g., they stil believed that it was important to
count the object adjacent to the previously counted object). This immature understanding
of the basic math skill of counting was found to be significantly correlated with the use of
immature strategies to solve addition nurnber facts. Geaq (1990) and Geary et al. (1991)
also noted that primary students with math disabilities took longer to move from wunting
al1 to counting on from the larger number in addition, suggesting that it may have taken
them longer to discover a d o r undentand the commutative property of addition. Ohlson
and Rees (1991) suggested that computation procedures are guided by conceptual
knowledge so that procedures will be modified to conform to the beliefs.
Most of the studies of basic math concepts have been carried out on students in
the elementary grades and have only looked at one type of basic concept. Therefore, an
overalt mode! of the development of math concepts has yet to be created. Zt appears that
students with math disabilities may have some delays in the development of al1 of these
concepts sbrting frorn the fiagetian conservation concepts, through to counting
principles, decomposing and recombining numben, place value concepts, properties of
operations and estimation. An understanding of place value seems to help with the
adaptation of procedures to new questions and the recall of procedural steps and d s o
reduces dgorithmic errors. The possession of estimation skills appears to reduce the
demands on cognitive processes such as rnemory, thereby freeing up these processes for
other aspects of the math question.

(iii) Number Facts


One of the more frequently occumng difficulties in mathematics is an inability to
acquire and maintain basic math facts at a level of automaticity that is sufficient for
acquiring higher level math skills (De Corte & Verschaffei, 1981; Fleischner, Garnett, &
Shepherd, 1982). Many children with math disabilities have this problem (Geary et al.,
1991; Goldman et al., 1988; Howeil et al., 1987). A number of researchers (Ashctaft,
Yamashita, & A r a , 1992; Geary et al., 1991; Goidman et al., 1988)have suggested that
the acquisition of number fact automaticity remains a persistent deficit for children with
math disabilities, but that their use of more immature counting strategies reflects a
developmental delay. The acquisition of number facts is one of the more frequentiy
studied areas of mathematics. It is also one of the few areas in which models have been
created and developmental patterns have been identified.
Number facts are the most basic set of data in arithmetic and îhey are much like a
sight vocabulary in reading. Both are isolated bits of information that can be used to
complete a whole (problem or sentence) and both are more useful if they can be
irnmediately recalled so that the individual can focus on higher level processing (problem
solving or complex computations in math and comprehension in reading) (Enright, 1985).
It is generally assumed that accuracy and fluency (high speed of performance) in basic
number facts are fundamental to proficiency in mathematics (Case, 1985;Fleishner &
Manheimer, 1997). Geary and Wideman (1987) found a strong relationship between the
speed of basic arithmetic processes (e.g., retrieving number facts and regrouping) and
success on numencai operation paper and pencil tests in college students. Atkinson
(1983) noted that students who took longer than two minutes to complete sixty
multiplication facts were at a disadvantage when they reach secondary Ievel mathematics.

Number Fact Strateees


Fuson et al. (1997) describe three levels of development in the mastery of addition
and subtraction number facts, but they also seem to apply to multiplication and division.
in Levei I, objects are used to count out al1 of the numbers in a question. In Level2,
students can add on in addition and count up or down in subtraction. In Level3, students
can use decomposing and recombining to transform the questions and thus use a derived
fact strategy. Nurnber fact automaticity is not identified as a separate level as
memorization of the facts cm occur at any of tbe above three levels.
The counting strategies in Levels 1 and 2 provide the opportunity for associating
the operands (2+3) with the correct answer (5) leading to the eventual memorization of
the number facts. However, if counting strategies are too slow or working mernory is
poor, the operands will decay before the answer is determinecl so that they are less likely
to become associated in long-term memory. When performing addition, the counting
strategies usualIy progress from counting d l (e-g-, 2 + 5 = [1,2]; [l, 2,3,4,5]; [ l , 2,3,4,
5.6, TI), to counting on (2 + 5 = 2,3,4,5.6,7) to counting on from the larger number
(min d e ) fe-g., 2 + 5 = 5,6,7).Counting on from the larger number would suggest that
students have understood the commutative property of addition (a + b = b + a). In
subtraction, children usually progress from only counting down (e.g., 7 - 2 = 7,6,5)
to
following the srnall count model, so that they count up or down depending on which
direction will require the least amount of counting (e-g., 8 - 2 = 8.7,6but 8 - 6 = 6,7,8).
If subtraction is taught as both take-away (10 take away 4 = ?) and how many more (4
and how many more? = IO), then both counting down and counting up strategies can be
used and subtraction quickly becomes as fast and accurate as addition (Fuson et ai.,
1997). However, if only the take-away interpretation of subtraction is taught, then
students must determine the counting up strategy on their own and a lag between
subtraction and addition mastery occurs (Fuson et al., 199'7). Two common counting
strategies in multiplication are repeated addition (e-g., 5 x 3 = 5 + 5 + 5) and counting by
n (e-,o.,2 x 4 = 2,4,6,8)(Geary, 19%).
Two more sophisticated back-up strategies that apply to al1 operations are the use
of rules (e.g., n x O = O. n + O = n) or derived facts (e-g., using a known fact to detemine
an unknown fact) (Geary, 1996). These two reference strategies c m also lead to the
development of number fact automaticity. When using a denved facts strategy, tie
problems (4 x 4). questions involving 5 and 10 or number facts from other operations are
often used as referent facts. Derived fact solutions in the United States often involve ties
(e.g., 7 + 8 = 7 + 7 + 1 = 15) (Fuson, 1997). whereas in Asian countries, the number ten
is used to reconstnict questions (e.g., 7 + 8 = 7 + 3 (=IO) + 5 = 15) (Fuson, Stigler &
Bartsch, 1986). This ten strategy is mainly used in the United States when nine's are
involved in the question (Steinberg, 1985).
There is the least amount of research on division number facts. It appears that to
solve division questions children tend to refer to their multiplication (e.g., for 15 + 3 they
rely on their knowledge of the 3 times tables, 3 x 2 = 6 , 3 x 3 = 9 , 3 x 4 = 12-3x 5 = 15)
or addition skills (e-g., for 15 s 3 they use repeated addition to detemine the quotient, 3
+ 3 +3 + 3 + 3) (Campbell, 1997).
Previous research has suggested that students use a combination of these counting
and reference back-up strategies along with automatic recdl (Bisanz & LeFevre, 1990;
Geary & Burlingham-Dubree, 1989; Goldrnan, Mertz, & Pellegrho, 1989; Lemaire &
Seigler, 1995). Cooney et al. (1988) found that grade 3 and 4 students used automatic
retrieval for 55%and 74% of the presented multiplication questions, respectively, with
repeated addition, derived facts and rules used on the remainder. Improvements in
number fact reineval speed and accuracy could result from improved execution of the
strategies seiected, changes in the mix of strategies (frequency each strategy is used and
the types of strategies used), or increased levels of automatic recdl (Lemaire & Seigler,
1995). Students generally make adaptive choices between whether to use a back-up or
automatic retrieval strategy, with the former used more often for more difficuit number
facts (Lemaire & Siegler, 1995: Siegler, 1987, 1988). Chronometric data seems to
suggest that by grade 4, children with average math achievement have switched from a
counting strategy to an automatic retrieval strategy for addition (Ashcraft & Fierman.
1982)- subtraction (Kaye et al., 19%; Kaye et al., 1989) and multiplication (Cooney et
aI., 1988).
Siegler (1986) developed a strategy choice model for addition. but it appears that
it can be generalized to other operations as well. This model suggests that two factors
influence an individual's strategy choice. The first factor is the associative strength
between the operands and the answer. If the operands are associated mainiy with one
answer, the distribution is said to be peaked, whereas if they are associated with rnany
answers, the distribution is described as flat. The more peaked the distribution the more
likely the question will be answered with automatic recall. The second factor is the
confidence that the student has in the correctness of hisher response. The confidence
criterion varies from individual to individual, thus the same cod5dence ievel rnay lead
some students to respond automatically and others to resort to a back up strategy.
The shape of the distribution can be influenced by a number of factors. The more
errors that are made retrieving a given number fact, the flatter the distribution. As back-
up strategies involving larger numbers seem to be more emor-prone, the distributions fcr
larger number facts are more likely to be flat (LeFevre et al., 19%). Siegler (1988) also
found that the slowoess of back-up strategies influenced the use of automatic retrieval.
finally. increased exposure to a number fact would increase the associative strength. but
interference from already leamed problems and problems involving the same operands in
other operations could flatten the distribution (Campbell & Graham, 1985).
Currently, there is some debate as to whether there is a cornmon retrieval network
for the number facts in reciprocal operations (e.g., addition-subtraction; multiplication-
division) or whether each operation is independent There is sorne evidence to support
both arguments in the neurupsychological literature on individuals with brain injuries or
lesions (e-g., Hitanair-Delazer, Semenza & Denes, 1994; Cipolotti & de Lacy Costello,
2995). The fact that individuals initially rely on addition to solve subtraction questions
and multiplication to solve division questions suggests there rnay be some commonaiity
(Geary, 1994; Siegler, 1988). Campbell (1997) also found that multiplication and
division had similar error patterns (both in the types of errors and the distance these
errors were from the correct answer) and reaction times. However, the lack of transfer of
practice effects between operations (Rickard, Healy & Boume, 1994) supports a more
independent retrievd process.
Number Fact Characteristics
It has been noted that retneving nurnber facts appears to become more difficult as
the operands get larger, which is referred to as the problem-size effect The problem-size
effect can be measured by response Latencies, erron or the number of different solution
strategies (Hamann & Ashcraft, 1986). Tie problems, w hich involve repeated operands
(e.g., 7 x '7) are an exception to this rule, as they are easier to solve than non-ties of
approximately equd value (e.g., 7 x 6) (Ashcraft & Battaglia, 1978; Graham &
Campbell, 1992). Tie effects have been found in ail operations, thus it has been
suggested that their ease is not sirnply because the repeated operands draw from one list
of answers (e.g., 7 x 7 o d y calls up the 7 times table ) (Campbell, 1997).
A number of different hypotheses have been proposed to explain problem-size
effects. In some children the problem-size effect can be attributed to their use of
counting strategies, which c m take longer and are more diffcult for larger problems,
with the exception of ties (Siegler, 1988). Studies have found that elementary textbooks
tend to use smailer-number problems more frequently (Hamann& Ashcdt, 1986;
Siegler, 1988), which results in less practice solving larger-number problems. It has also
been proposed that because iarger-number number facts are usually introduced later,
cumulative proactive interference could make them more diffrcult to rnemorize
(Campbell & Graham, 1992).

Number Fact Errors


Sign errors c m be made in al1 four operations (e.g., doing addition for
multiplication) and most commonly occur when ties are involved (e.g., 4 x 4 = 8)
(Campbell, 1997). Other error patterns are generaily unique to each operation, but in
every case, each time an error is made, an incorrect association is reinforced, which leads
to retrievai problerns (Geary, 19%). The most common nurnber fact error in addition is
undercounting by one (Geary, 1990; Siegler & Robinson, 1982). This often occurs
because the cardinal value of a number is counted twice (e.g., 4 + 3 = 4,5,6).A similar
pattern has been found in multiplication, where e m r s are usually one operand away from
the correct answer (e.g., 6 x 7 = 36 is correct if the 7 is reduced by 1 creating 6 x 6 = 36)
(Campbell, 1997).
Two types of errors are commonly found in the multiplication of single-digit
numbers. A table-related error is identifiable because the answer is in the correct
multiplication table, and is thus a multiple for oniy one of the operands in the question
(e.g., 3 x 8 = 21 is a table-related e m r as 21 is a rnukiple of 3). Usuaily the errors relate
to the smaller of the operands (Campbell, 1997). These errors occur most frequently
when both operands are associated with the same incorrect answer (doubiy related errors)
(e.g., 4 x 8 = 24,2 x 6 = 18.3 x 9 = 18,6 x 9 = 36) (Campbell & Graham, 1985). Table-
related emrs are said to be operand-driven (sometirnes called factor-dnven) and are the
most comrnon of multiplication errors (Campbell & Graham, 1985; Siegler, 1988).
Children who use either repeated addition or counting by n strategies often add too many
or too few of the numbers and are prone to making table-related errors (Graham &
Campbell, 1992). Table-related errors can also be identified in division w ith questions
involving the double related errors (e.g., 184=2). More table-related division errors
occur when the divisor is small.
In the second type of error, near-misses, the magnitude of the correct answer is
determined by the combination of the operands, so there is a tendency for errors to be
numericaliy close to the correct answer ( ~ Wsmaller
O or larger than the correct answer),
such as 6 x 4 = 23 (Siegler, 1988). These types of errors reflect problem-driven
processing (Graham & Campbell, 1992). Near-miss erron are also present in division
(CampM l , 1997).

Number Facts and Math DisabiIities


Much of the research on number facts has focused on how number facts are
leanied as opposed to comparing differences between MD and NC Groups. For these
studies, number fact strategy selection has been determined by using response latency
times (chronometric data), observation or in a few case by interviewing the chiid.
Cooney et al. (1988) reported high correlations between verbal reporting of strategies and
latency data. Other researcher's have demonstrated that interviewing the child about the
strategy they used to answer a question provides the clearest information (Ericsson &
Simon, 1984; Fieischner & Manheimer, 1997; Siegler, 1989).
Geary (1990) looked at addition skîlls in grade 1 students and found that children
with math disabilities used more immature strategies (e-g., counting ail as opposed to
counting on), made more errors with ail strategies and had unusual time patterns for fact
retrievai. By grade 2, the NC students had made the expected shift to increased reliance
on automatic retrieval. In contrast, the MD children were rarely using automatic recall,
but they had shifted from counting al1 to counting on. The MD Group continued to make
a high rate of errors and have unsystematic retrieval tirnes. Studies by Geary, Brown and
Samaranayake ( 1991) and Geary, Bow-Thomas and Y ao ( 1992) have aiso found that
students with MD tended to use immature counting strategies (e-g., ,counting al1 or
counting on from the first number as opposed to counting on from the larger number).
Using chronometric data to compare addition number facts in grade 2,4 and 6
students, Geary et al. (1987) once again found a shift in the strategy mixes of the NC
p u p , from rnainly counting in the grade 2 students to maidy automatic recall in the
grade 6 students. However, the LD students coutinued to rely on counting strategies in
dl of these grades. They also noted that in the second grade, the LD Group was slower at
executuig the counting strategies than the NC Group and had a higher error rate (13%).
Geary (1996) also suggested that students with MD make poor strategy choices (not
selecting the strategy that will most efficientIy give them the correct answer).
The research on students with MD indicates that they display deficits in number
fact automaticity relative to normally achieving p e r s and they also have a higher
proportion of errors when an answer is retrieved autornatically (Fleischner, Garnett, &
Shepherd, 1982; Gamett & Reischner, 1983; Goldman, Pellegrino, & Mertz, 1988). If
the lack of automaticity is due to lack of experience (Nt&& McAuley, 1991)then
practice should lead to improvements. Some support for this was demonstrated by Geary
and Brown (1991) as they found that experience led to an increase in the use of direct
retrievai for number facts and fewer computationai errors, but slow retrievat did not
disappear. Goldman, Pellegrino and Mertz (1988) carried out an interesting study
Iooking at the benefits of practice on the acquisition of addition number facts. They
determined that with practice there was an increased reliance on automatic fact retrievai
in students who were already using automatic retrieval a fair amount. For the students
using a mixture of strategies, there was a shift towards using the more sophisticated back-
up strategies and some increase in automatic retrieval. Those students using basic
couoting strategies became more accurate, but litîle shift in their strategy mix occurred.
This findings suggests that there are different effects of practice depending on what stage
of number fact acquisition a student has reached. Therefore, it appears that studies
assessing the benefits of practice on students with math disabilities should include more
than just measures of automaticity or they may miss the more subtle changes in strategy
mix. Rickard, Healy and Boume (1994) tested university students and found that there
was transfer of practice between multiplication questions with opposite operand order
( e g , 3 x 4 and 4 x 3) but not between division problems with opposite operand order
(e-g., 12+ 3 and 12 + 4) oor between the two operations, despite their reciprocai nature.
The research on oumber facts suggests that there are a number of stages that an
individual progresses through as hdshe develops number fact automaticity. A number of
factors affect the students ability to transfer from a back-up strategy to automatic
retrievai, ranging from exposure to a question, tie effects, problem-size effects and
number of errors to confidence level. Along with these factors, researchers have
generally associated poor number fact automaticity with memory problerns. particularly
in working rnemory. Researchers believe that students with MD tend to display a deficit
in their ability to retrieve number facts automatically and a delay in their development of
more mature back-up strategies. With most of the number fact studies investigating only
one operation, the evidence needed to support a cornmon or separate network retrieval
model across operations has yet to be collected.

(iv) Cognitive Variables


Mathematics includes a wide variety of topics and skills, with different cognitive
processes required to perforrn each aspect of this subject area. It is also possible that
cognitive deficits may be relevant at one age, but not at another (Ginsburg, 1997). Thus
it is udikely that there is one single cognitive deficit that causes math disabilities. The
cognitive variables that have been investigated in math computation research are working
memory, short-term memory, long-term mernory, processing speed, sequencing skills and
visual-spatiai abilities. Geary (1993) has also mentioned attention allocation as a factor.
With a few exceptions, studies investigate each of these co,g&ive areas in isolation so
that no overall profile of students with math disabilities has been developed. The
contribution of each of these areas to math computation is also rarely addressed as rnost
studies do not include regession analyses, but only compare NC and MD students to
each other.

Memory
Mathematical taslcs require the use of strategies to store new information and the
activation of already stored material from long-term memory (Swanson, Cochran &
Ewers, 1990). The combination of storage and processing functions is cdled working
memory (Baddeley, 1981). Baddeley suggests that working rnemory is composed of
three components: a central executive, which is the conîrol system that oversees the
selection and operation of different processes, an articulatory b o p for verbal memory and
storage, and a visuo-spatial scratch pad for imagery and spatial memory. In anthmetic it
is assumed that the articulatory loop stores the numbers and sign of a question, while the
central executive retrieves the algorithm and number facts for the problern (Logie,
Gilhooly, & Wynn, 19%; Siegel & Linder, 1984). Therefore, a poor working memory
wuld hinder the successful completion of computations, especially those that are more
complex or involve less weii known number facts.
Working memory is thought to be important in the initial stages of number fact
mastery as the operands must be temporarily stored while the answer is being determineci,
usually using a counting strategy. Studies investigating number fact knowledge suggest
that children with arithmetic difficulties have poor working memories which lead to
incomplete representations of number facts in long-term memory (the operands do not get
associated with the answers) (Geary, 1990; Geary, Bow-Thomas, & Yao. 1992; Geary &
Brown, 1991; Hitch & McAuley, 1991; Siegler & Shrager, 1984). As a result, children
with math disabilities tend to continue to rely on the more time consuming back-up
strategies for number fact retrieval which takes up much of their limited working memory
(Ackennan, Anhalt, & Dykman, 1986; Hasher & Zacks, 2979). Working mernory has
also bsen implicated in the development of early counting strategies as Geary,Bow-
Thomas and Yao (1992) found that children with better working memones used more
verbal as opposed to fioger counting strategies. Investigations of working memory and
math have involved students with dficulties exclusively in arithmetic. The findings
suggest that deficits in this group are only present when numbers are involved (Hitch &
McAuley. 1991; Siegel & Ryan, 1989).
A number of studies, but not dl,have found that children with poor arithmetic
skills have some form of short-term memory (STM) defitit A few studies have found
that these memory deficits are pervasive (Swanson, 1993;Turner & Engle, 1989),but in
generai the content of the memory tasks mems to be important. Some researchers have
suggested thaî for those who are good at math, digits have a stronger representation so
that they are identifiai more rapidly, resuiting in betîer memory performance on tasks
involving digits (e-g., Case et ai., 19û2). Dark and Benbow (2990) found that
mathematically gifted adolescents did significantly better on tasks hvolving digits and
spatial locations whereas verbally gifted children did better on tasks involving word
stimuli. This finding is consistent with most of the results from memory studies
involving snidents with math disabilities. Fletcher (1985)found that students with math
disabilities did worse on a visual STM task, whereas those who had a reading disability
did more poorly on a verbal STM task. Webster (1979) also detennined that grade 6
children with math disabilities did signifxcantly worse on STM tasks than normal controls
and that mode of presentation was a factor, but the reading levels of the participants were
not identified. When Bull and Johnston (1997) controlled for reading no STM effects
were found Ackerman and Dykrnan (1995) wmpared elementary students with reading
disabilities to those with reading and arithmetic disabilities based on their performance on
the WRAT-R. They also found no simcant differences between the two groups on any
of the auditory or visual memory span measures. They reported similar results with a
sample of adolescent students.
Speech rate is often associated with short-term memory (Baddeley, Thomson &
Buchanan, 1975; Ellis & HennelIey, 1980; Hitch, Hallidrty & Littler, 1989). Baddeley
(1986) suggested that in arithmetic, speech rate should be related to counting speed. The
faster the counting speed, the longer the short-temi memory span for numbers (Kail,
1992), because fast articulation means more items can be rehearsed in memory (Cowan,
1992; Cowan et, al., 1994). However, Hitch and McAuley (1991)found that although
children with specific arithmetic difficuities performed poorly on short term memory
tests. their articulation rates were not significantly slower. Therefore, they and others
have suggested that item identification (accessing items from long-term memory) is a
factor in STM,with improvements in the former leading to increased capacity in the
latter (Bull & Johnston, 199'7; Case et al., 1982; Ericsson & Kintsch, 1995; Gathercole &
Adams, 1994; Henry & Miller, 1991; Hitch & McAuley, 1991; Hulme et al, 1991;
Roodenrys et al, lm).
Long-term memory (LTM) is involveci in the recall of number facts and
procedural steps. Each time a number fact is encountered an association between the
operands and the answer is formed and it is stored in LTM. Students with math
disabilities seem to have trouble recalling number facts from LTM automatically
(Fieischner, Garnett, & Shepherd, 1982; Goldman, Pellegrino, & Mertz, 1988). Geary
(1993) has suggested that the CO-occurrenceof reading and arithmetic difficulties may
reflect poor retrieval of semantic information from long-term memory.
Sequencing difficulties have been proposed as a potential area of weakness in
children with math difficulties. Trouble learning sequences may affect counting fluency
(Geary et al., 1992; Hitch & McAuley, 1991) and lead to procedural deficits (Geary,
1990, Geary, 1993). Buil and Johnston (1997) foumi that alphabet and number
sequencing were done more poorly by students with math weaknesses.
Thus, the findings from previous research seems to support the idea that students
with MD often display deficits in working, short-term, long-term and sequential memory,
particularly if' the task involves digits or spatial content. These different types of memory
are thought to be important for both the number fact and procedural aspect of
computation.

Processinp S ~ e e d
It has been suggested that slow execution of operations is the major problem for
children with arithmetic difficulties (Gametî & F'ieischner, 1983;Geary, 1993). This
could be due to generally slower processing speed abilities andlor a lack of automaticity
in the basic facts which hinders performance on more complex mathematical tasks.
There are very few mathematics studies that have included measures of processing speed.
Ackeman and Dykman (1995) found lower scores on the Wechsler Intelligence
Scale for Children - Third Edition (WISC-III) (Wechsler, 1991) Coding and Symbol
Search subtests in adolescents with poor arithmetic and reading (AR Group), compared to
students with ody poor reading skills (R Group). However, this finding reflected in part
the unequal sex distribution, as gids in their study scored higher than boys on these two
subtests and there were proportionally fewer girls in the AR Group. Ackennan and
Dykman (1995) also conducted a regression analysis using the Wechsler Intelligence
Scale for Children - Revised (WISC-R) (Wechsler, 1974) subtests as predictors of the
WRAT-R fithmetic scores of 65 elementary school children with reading and AR
disabilities. The Arithmetic, Infoxmation and Block Design subtests predicted alrnost
hdf of the variance in the scores. m e n the Arithmetic subtest was not dlowed to enter
the stepwise regression until last, the information,Block Design and Coding subtests
were the top three predictors respectively, also accounting for about half of the variance.
Bull and Johnson (1997) reported poor performance on both a visud oumber
rnatching task and a pegboard test in a group of elementary students whose math scores
were below the mean of the group when compared to students above the group mean.
This is one of the few shidies to look at the relationship between cognitive variables and
math scores. They found that processing speed in elementary students explained 8%of
the variance and was the only factor aside from reading (45%of the variance) that made a
significant contribution to the regression formula. Bull and Johnson (1997) also noted
that processing speed was siWcantly correlated with addition counting enors and
response reûieval time.
A shidy by Benbow and Minor (1990)comparing rnathernatically precocious 13
year old students with verbally precocious p e r s found that the former did significantly
better on measures of spatial skills, nonverbal reasoning (Raven's Progressive Matrices
processing speed abilities (Differential Aptitude Test
[Raven, Court & Raven, 19m),
Clerical Speed and Accuracy subtest Bennett, Seashore & Wesman, 19741) and what
they considered memory (WISC-R Coding subtest).
The results from these studies suggest that students with MD generally do not do
as wetl as their peen on measures of processing speed and that it may be a factor in math
computation.
(vi) Math Disabilitg Subtgpes
For most of the research in mathematics, MD students have been treated as a unit.
However, neuropsychological studies have provided insight into the potential
heterogeneity of children with math disabilities. Rourke and his colleagues (Hamadek &
Rourke, 1994; Rourke, 1993; Rourke & Conway, 1997; Rourke & Finlayson, 1978;
Rourke & Strang, 1978; Strang & Rourke, 1983)have proposed that subtypes of children
with poor math skills can be identifed based on their academic profiles on the wde
Range Achievement Test (WRAT) (Jastak & Jastak, 1%5). Their series of studies
involved 9 to 14 year old children with Average intelligence drawn from a large clinical
sample of children seen for l d n g or perceptual problerns. In order to facilitate
understanding, the same labels are used for each subtype, even though individual studies
may have used different unes. One subtype consists of children with unifody poor
math, reading and spelling skills (Group ARS). A second subtypes includes students
with low scores in al1 three academic areas, but whose math was betîer than their reading
and spelling (Group A>RS). The final subtype they discussed consists of children with
poor math skills but average reading and spelling skills (Gzoup A). Rourke and his
colleagues used grade quivalent and percentile scores to identify the subtypes, making it
difficult to determine how these criteria would be defined with standard scores.
In each of the studies c b e d out by Rourke and his associates, cornparisons were
made between the subtypes using different neuropsychological measures. Rourke and
Finiayson (lW8) found that the ARS and AAtS Groups did simcantly better on
measures of visual spatial abilities (e.g., WISC Block Design subtest) resulting in a
higher WISC Performance than Verbal Scale score. In contrast, the A Group was
superior on verbal tasks, including rote verbal memory (e.g., WISC Vocabulary and Digit
Span subtests) which redted in the opposite pattern, a higher WSC Verbal than
Performance Scale score. Rourke and his colleagues are not alone in associating math
difficulties with poor visual spatial organization (Strauss & Lehtinen, 1947; Johnson &
Myklebust, 1%7; Willows, 1991). Further studies indicated that the Group A students
also had deficiencies in psychomotor and tactile-percepnial skills (Rourke & Strang,
1978), nonverbal prublem solving (Strang & Rourke, 1983) and visual-perceptual-
organization (Hamadek and Rourke, lm)compared to the other two subtypes. Based
on Group A's pattern of neuropsychologicai strengths and weaknesses the students in this
group were identified as having a nonverbai learning disability (NLD) (Harnadek &
Rourke, 1994) and their difficulties were related to a right hemisphere dysfunction
(Rourke. 1993). The poor math skills of the Group A students with the NLD profile are
thought to be the result of their underlying neuropsychological problems, particularly in
the visual-spatial areas. Ozols and Rourke (1988) found a similar pattern of results in a
study involving 7 and 8 year olds from di three subtypes, but the findings with these
younger students were not as clear cut.
A number of other researchers have tried to replicate the findings of Rourke and
his colieagues with mixed success. Branch, Cohen and Hynd (1995) did not find that
right hemisphere dysfunction, identified by a lesion or a set of criteria associated with
disorders in each hemisphere, was more frequently associated with the Group A pattern
of arithmetic scores below the reading and spelling scores on the WRAT. Ackerman,
Dykman and Peters (1976)found no significaot differences in the WISC verbal and
performance profiles of adolescents with A, ARS and AS (low scores in arithmetic and
spelling, average score in reading) profües on the WRAT.
Share, Mofft and Silva (1988) tried to replicate the findings of Rourke and his
colleagues using subjects from the general population and including a normal control
group. They also looked at boys and girls separately. Unlike Rouke and his colleagues,
they used the Progressive Achievement Tests of Reading Comprehension (Elley & Reid,
1949) and Mathematics (Reid & Hughes, 1974)to identify which students belonged to
the A, AR or NC Groups. They clustered the masures they used into language and non-
language based categones in order to determine if the A and AR Groups would display
the expected cross-over effect (Group A would have language > non-language,Group
AR the opposite pattern of non-language > language). For the boys, they found the same
pattern as Rourke, as Gmup A scored significantly higher than Group AR in the language
area and significantly below Group AR in the non-language area. However, this cross-
over pattern did oot occur for the girls. Group AR girls çhowed a similar pattem to
Group AR boys, but Group A girls were not sibgnificantly different than the NC Group.
The inclusion of a NC Group enabled Share et al. (1988) to determine that even the
strengths of the math subtypes were below the level of their normally achieving peers and
thus may also contribute to their arithrnetic difficulties.
Nolan, Hammeke and BarHey (1983) compared Group A students to a NC Group
with sirnilar reading levels and found no sigaifcant differences on the verbal or oon-
verbal measures of the Luria-Nebrasks test battery. However. they did not correct for the
fact that the Group A students were slightly older. In a study of memory abilities,
Fletcher (1985) found that Group A had difficuities with nonverbal memory tasks, but not
verbal ones, whereas Group AR had trouble with both. Siegel and Ryan (1989) reported
that working memory difficulties were o d y present in Group A if numbers were
involved, whereas Group AR had difficulty with both numbers and words. Hitch and
McAuley (1991) noted the same pattern in their study, but also determined that Group A
had lower digit span scores and were slower counters (e-g.. counting from 1 to 20).
Shafrîr and Siegel (1994) compared, NC, A, R and AR Groups on a number of
cognitive variables, but they did not covary out the group differences in estimated IQ
(NC > A, R > AR). which reduces the interpretability of their findings. The R and AR
Groups were found to have deficits in short-term memory on the digit span, rhyming and
nonrhyming word tasks in cornparison to the A and NC Groups. This is consistent with
fmdings from other studies that have found short-tenn mernos, deficits in students with
reading disabilities (Siegel & Linder, 1984; Siegel & Ryan, 1988). The A and AR
Groups displayed visual-spatial deficits, but for Group A. these were only present at the
post-secondary level. No memory deficits were found in Group A. In general. Shafrir
and Siegel (1994) found that the AR Group had lower mean scores than the NC and A
Groups on most tasks and the R Group on some tasks, which the authors suggested meant
that the AR Group was the most severely learning disabled of the three groups.

Arithmetic Patterns bv S u b t v ~ e
Rourke suggested that the different subtype profiles would lead to different types
of computation errors. He examined the WRAT Arithmetic protocols of the AIRS and A
Groups in order to determine if the former made more errors related to their verbal
deficiencies and the latter made more spatial errors. Rourke (1993) reported that the
expected patterns did ernerge. The A>RS Group had trouble with nurnber fact recali and
made errors in mechanical math due to verbal deficiencies. He also noted that they
would only attempt questions they knew how to do. In contrast, the A Group
demonstrateci poor reasoning andjudgement by trying questions they did not know how
to do, made spatial errors (e.g., misaligning numbers, directiondity confusions), visual
detail errors, graphomotor errors. missed or added steps to procedures, failed to shift
psychologicai set (e.g., continued using the same operation for a third question even
when the sign changed) and to a lesser extent made number fact errors.
Saito (1992) used a modif~edversion of Englehardt' s ( 1977) classification
categones when comparing computation errors on the WRAT-R in grade 3 and 5 students
with and without math difficulties. She found no qualitative diffe~ncesin the erron of
the Gmup A and Group AR students, with both MD subtypes making mare errors than
the NC Group.
Jordan and Oettinger Montani (1997) compared grade 3 studenîs in Groups A. AR
and NC on number fact tests. They found that Group A was wone than the NC Group in
timed, but not in untimed conditions. Group AR did worse than the NC and A Groups in
both conditions. They noted that both Gmup A and AR relied more on back up strategies
to solve number fact questions than the NC group, but Group A executed them more
effectiveiy than Group AR which ailowed them to be as successfd as the NC Group in
untimed situations. The NC and A Groups were found to be more accurate than the AR
Group with automatic retrieval. The NC h u p also made fewer errors on back up
strategies than the AR Group, but not the A Group.
The majority of the math subtyping literature has been produced by Rourke and
his colleagues using students from clinical populations. When subtypes are found,
students w ith a specific deficit in arithmetic display visual-spatial weaknesses, whereas
students with more global academic weaknesses display verbal difficulties. The Group A
and A>RS profiles also seem to be associated with different types of computation emors.

Limitations to Previoas Research


There are a number of limitations to the math research that has been carried out.
One of the main problems is the use of widely varying criteria to identify children wiîh
math disabilities. A nurnber of tesearchers have used a cutoff score of less than the 25fh
percentile on the Arithmetic subtest of the W~deRange Achievement Test (WRAT) to
identify a math disability (e.g., Ackerman, Anhalt, & Dykman. 1986; Siegel & Ryan,
1988;Swanson, 1994). Others have used a cutoff point of at or below the 3 6 percentile
on the WRAT (e-g., Fietcher, 1985)or the Progressive Achievement Tests (e-g., Share,
Moffit, & Silva, 1988)or below the 4@ percentile on the Woodcock-Johnson Tests of
Achievement - Revised (WJTA-R; Woodcock & Johnson, 1989)(Geary, Bow-Thomas,
& Y ao. 1992). Some investigators have used the sampie mean to divide their subjects
into low and high math groups (e.g., Bull and Johnston, 1997; Rasanen & Ahonen, 1995),
whereas others used median splits (e.g., Ackemian & Dykman, 1995). finaily, sorne
math disabilities have been identified using nurnber of grades below placement (e.g.,
Rourke & Finlayson, 1978; Rourke & Strang, 1978; Webster, 1979) or a specially
developed formula based on achievement age and expectancy age (McLeod and Crump,
1978). The variability in the cut-off criteria indicate how arbitmy the identification of a
disability becomes when scores are continuous.
Although there are a few studies th& do not hclude intelligence cnteria (e.g., Buil
& Johnston, 1997; Webster, 1979), the majority of studies have a minimum score
requirement on a mesure of intellechml potential (e-g., the Verbal, Performance or Full
Scales of the WISC, the Peabody Picture Vocabulary Test - Revised [PPVT - R] [Dunn &
Dunn, 19811). However, the minimum requirement varies fmm a standard score of80
(e.g.. Ackerman & Dykman, 1995; Shafnr & Siegel, 1994), to 85 (e.g., Fletcher, 1985;
Rasanen & Ahonen, 1995: Rourke & Strang, 1983) or 90 (e.g., Share, Moffit & Silva,
1988).ALI of this variation in criteria for identifying math disabilities makes it more
difficult to generalize the findings.
Much of the research has been conducted with elementary school children (e-g-,
Geary, 1990; Rasanen & Ahonen, 1995; Russell & Ginsberg, 1984) so îhat the findings
may not apply to older studeots or to the components of mathematics covered in higher
grades. Many studies also involve groups with wide age ranges ages (e.g., Ackeman &
Dykrnan, 1995; Andersson. Richards, & Hallahan, 1980;Rourke & Strang, 1983; Siegel
& Ryan, 1988)which may mask some of the findings as different abilities may be
required tu meet task demands at different ages. A few of the studies also use faidy srnaIl
sample sizes (e.g., Hiebert & Weame, 19%; Rourke & Strang, 1983).
Finally, students are identified as having math disabilities based on their scores on
mathematics tests, but these studies do not include critena to identify which students are
actually "teaching disabled". Many elementary school teachers seem to be
uncornfortable with math and teach it poorly (Ginsburg, 1997). The language of math
used by teachers may contribute to some of the difficulties (Tracy, 1994). For example,
subtraction is often taught using the statement, "remember, you can't take a bigger
nurnber away from a smaller number". This is not true. as the answer can be a negative
number, and it may tead to some of the regrouping erron seen later. Streamlining the
identification of students in order to determine those who are truly MD is likely to
continue to be a challenge.

Ovewiew of Previons Findings


From the review of iiterature it becomes apparent that almost dl of the math
research that has been canied out investigates a specific area related to mathematics in
isolation. Therefore, there is no overall theoretical framework for the development of
n order to obtain a more global picture of computation, it is
math computation. l
necessary to combine the findings from individuai snidies. The compilation of previous
results seems to suggest a framework like the schematic drawing displayed in Figure 1.
Computation seems to be composed of three main parts: symbol recognition. number fact
knowledge and procedural knowledge. These three areas are represented in the second
row of the schematic drawing. The research findings to date suggest that students with
MD have particular difficulty with two of these areas, the acquisition of number fact
skills and the mastery of procedural steps. Difficuity with the former is thought to
represent a developmental deficit, whereas difficulty with the latter is thought to
represent a developmentai delay.
Each of these three areas are dependent on underlying concepts and coepitive
processes, which are represented in the fint row of Figure 1. Working, short-terni, long-
term and sequentiai memory are thought to influence the development of number fact
acquisition and the ability to retain procedurd steps in the correct order in memory while
they are king performed. Short- and long-term memory would also seem to be
important for symbol recognition. Some studies, but not d l , have shown that students
with MD have deficits in these types of memory compared to NC students. In most cases
the memory dficulties were confrned to tasks that included numbers. Processing speed
has also been related to both number fact acquisition and the performance of the
procedural aspect of computation. Although few studies have investigated processing
speed, it appears to be a potential area of weakness in students with math disabilities.
Basic math concepts seem to be important for developing an understanding of
computation procedures so that fewer errors are made and strategies can be generalized
from one type of question to similar questions. The level of understanding of basic math
concepts may also affect the selection of back-up strategies for number facts. Studies
with elementary students suggest that students with MD develop these basic concepts
more slowly than their NC peers.
It seems intuitive that visual-spatial skills would be related to symbol recognition
and number a l i m e n t in procedures. Deficits in the visual-spatial areas have pnmarily
been found in students with specific arithmetic disabilities (Group A) and seem to be
related to sign and alignment errors. Two areas that have rarely or never been researched
in students with math disabilities are number patterns (e.g., 2 + 10 = 12, 12 + 10 = 22,Z
+ 10 = 32, etc.) and attention. It is thought that the former would be related to number
fact acquisition, w hereas the latter would be important for all aspects of computation in
order to reduce errors due to inattention.

Rationale for the Stndy


There has been an increasing number of stuâies investigating the area of
mathematics and math disabilities, however, many gaps still remain. It is believed that
this is the only study to include rneasures of basic math concepts, number facts, rnemory
and processing speed al1 together and to relate these variables to math computation. This
snidy was somewhat exploratory in nature because some of these areas have rarely been
investigated in this age group. The rationale for including al1 of these different areas was
to provide a broader understanding of which aspects of math computation were difficult
for grade 7 and 8 MD students and what factors may contribute to these difficulties. It
was hoped that this could then become a startinp point from which to build a unaied
theory of math computation.
This is the first study to include number fact questions for al1 four operations.
This provided the opporhmity to make cornparisons among the four operations in terms
of ease of automaticity, difficulty ievel of individual number facts. error patterns and
retrieval strategies. As a result, the applicability of Fuson's levels of number fact
development to al1 operations, and the possibility of a common reûieval system could
both be addressed.
In many previous studies, the MD Group has been composed of students with a
wide range of skills in other academic areas. Therefore. it was unclear whether the
results applied to al1 of the students in the MD Group or whether some of the potential
findings were masked because of the possible heterogeneity of the MD Group.
Therefore, in the present study analyses were carried out with the MD Group as a whole
as weil as with the A and ARS subtypes from this MD Group. It was believed that this
design would provide information about the similariries and differences between MD
subtypes and would also indicate how weli the MD Group represented these subtypes.
The results from this study could aiso provide guidelines for the creation of remedial
ProgramS.
Foilowing the assumption that the grade 7 and 8 MD students in this study were
similar to those found in studies of younger MD students, the presence or absence of
difficulties that other researchers had noted in elementas, MD students, may shed sorne
light on whether the MD Group's computation difficulties represented a developmental
delay, deficit or difference. Differences between the grade 7 and 8 shidents could also
provide some information on this topic. It is reco,@zed that studies investigating delays
or deficits in students usually follow the same group of students over a period of time or
indude a younger control group with skills matched to the level of the MD Group, as
well as an age-matched control group. However, given the exploratory nahue of this
study, it was believed that although these findings should be interpreted cautiously, they
may provide some direction for fuhlre studies.
Figure 1
Schematic Representation of Computation Based on the Literature
1

Working Processing Number Attention Basic Concepts Sequential STM Visual-


Memory Speed Patterns (place value, Memory LTM Spatial
estimation) Skills

Nurnber Facts Procedure Symbol Recognition


METHOD

The fint goal of the design of this study was to identif'y grade 7 and 8 students
with math computation disabilities (MD Group) and average computation skills (NC
Group) in order to compare them on a number of dependent variables thought to be
related to cornputation. The second goal was to idenrify subtypes within the MD Group
in order to determine if students with difficulties ody in computation (Group A) differ on
the dependent variables from students with difficulties in computation, reading
(decoding) and spelling (Group ARS). The grade 7 and 8 age range was selected because
according to the Ontario cumcuhm guidelines. by the intermediate grades students have
been taught how to complete computations involving al1 four operations. Therefore, the
chance that students had cornputation difficulties solely because they had yet to be taught
a procedure was reduced.

SPbiects
The subjects were drawn from thirteen grade 7 and 8 classes at two senior public
schools in the Greater Toronto Area. A total of 418 grade 7 and 8 students (ages 12 - 14
years) were invited to participate in the study. No students were excluded from the study
due to intellechial or behaviour difficulties or their involvement in special education
classes. Of the 268 students who returned the letter of consent (64% return rate), 194
agreed to participate in the study. Due to the fact that some students were away on the
days that the Stage 1 group testing took place, 181 actually took part in the study. Of
those, 118 participants were from grade 7 and 69 were from grade 8.
The students al1 participated on a volunteer basis. A wntten consent fonn (see
Appendix A) was signed by the participants' parentlguardian and by the participant. A
letter accompanying the consent f o m outlined the purpose of the snidy, ensured subject
confidentiality and permitted withdrawal from the study at any time.
l?m!z&x
This study was divided into two stages. In Stage 1, the 187 participating students
were group administered the Arithmetic and Spelling subtests of the Wide Range
Achievernent Test - Third Edition (WRAT3) as weU as number fact tests for each
operation; a s part of a larger test battery. The total Stage 1 testing time was
approximately 40 minutes. The WRAT' subtest scores were used to determine which
students would be involved in Stage 2 of the study .
Those students with Arithmetic standard scores of 89 or below (a"
percentile)
were potential members of the math disability (MD) group and participateci in Stage 2.
The use of a WRAT3 Arithmetic cutoff score of 89 was consistent with the literature
(e-g., Ackerman, Anhalt, & D y h a n , 1986; Shafrir & Siegel, 1994: Siegel & Ryan, 1989)
and permitîed cornparisons between the current and past findings. No reading and
spelling cnteria were set in order to ailow for the identification of math subtypes.
Students with Arithmetic standard scores between 100 and 115 (50&to percentile)
and Spelling standard scores between 90and 110 (2Sb to 7 9 percentile) were aiso
involved in Stage 2 as potential members of the normai controi (NC)group. The
inclusion criteria for the NC Group were setected in order to try and ensure that these
students had solidly average math and cognitive abilities and no major difficulties in the
reading and spelling areas. The minimum WRAT3 Anthmetic standard score of 100 also
created an eleven point spread between the MD and NC Groups which was done to
reduce the overlap between the groups that might occur due to the standard error of
measurement.
Using these criteria 82 students were selected to participate in Stage 2, including
37 potential MD Group students and 45 potential NC Group students. A breakdown of
the results of Stage 1 based on the WRAT3 Arithmetic and Spelling subtest scores is
presented in Table 1.
Table 1

Stage 1Gmapings Based on WRAT3 Arithmetic and Spelling Standard Scores


Arithmetic Spetling Number of Students Continue to Stage 2
1
-l

s89 49 20 1 Y es - MD Group
J
I
I

439 >9C) 17 Yes - MD Group

I 1
100-115 90-110 45 Yes - NC Group

In Stage 2, the students were seen individually by the investigator in a small, quiet
roorn in each of the schwls. The testing began with the WRAT3 Reading subtest as well
as the Wechsler Intelligence Scale for Children - Third Edition (WISC-III) Vocabulary
and Block Design subtests. The results from these measures were used to further refine
the groups. Sîudents in the MD Group needed to have either a Vocabulary or BLock
Design scaled score between 8 and 13 to be designated as part of this group. This
flexibility would allow for visual-spatial versus verbal cornparisons amoog the math
subtypes. It also reflects the fact that individuais with learning disabilities often display
strengths in one scale over the other. The students in the NC Group needed a WRAT3
Reading subtest standard score between 90 and 110 ( p
to 75" percentile) as well as
both the WSC-III Vocabulary and Block Design scaled scores between 8 and 13. AI1
students needed to have an estimated Full Scale Intelligence Quotient (FSIQ) standard
score on the WISC-III of 80 or above. The Vocabulary and Block Design subtests were
used to determine the short f o m estimate of the FSIQ score. The reliability and validity
coefficients of this short form are .91 and .86 respectively (Sattler, 1992).
The students in the MD Group were further subdivided into subtypes based on
their WRAT3 subtest profdes following the criteria used by Seigel and Heaven (1986),
Shafnr and Siegel ( 1994) and Hmadek and Rourke (1994). Thus, students with
WRAT3 standard scores of 89 or lower (ad
percentile) on the Anthmetic subtest and
92 or higher (r3 0 percentile)
~ on Reading and Spelling subtests were identified as Group
A because their difficulties were specific to Anthmetic. The Group A members had
Anthmetic standard scores that were 10 or more points below their Reading or Spelling
scores (range from 11 to 27),which is consistent with the Hamadek and Rourke
methodology ( 1994).
Those students with scores of 89 or below (a"
percentiie) on dl three subtests
of the WRAT3, suggesting that they had deficits in arithmetic, reading and spelling, were
identified as Group ARS. The Anthmetic subtest scores of the Group ARS subjects were
within 7 points of their Reading or Spelling subtest scores. ln three cases the Arithmetic
subtest score was below the Reading and Spelling subtest scores, in five cases it was
above and in the remaining thne cases i t was in between.
Five students were found to have WRAT3 Arithmetic and SpeUing standard
scores of 89 or less (a"
percentile) and Reading standard scores of 92 or more (z30&
percentile) (Group AS). However, due to the small sample size, the AS Group was not
included in the analyses. The remaining three students did not meet any of the subtyping
criteria. The MD, NC, A and ARS Group requirements are presented in Table 2.
Of the eighty-two students who began Stage 2, twenty-two were eliminated as
they did not meet the above criteria for group membership. Seven of the thirty-seven MD
students were lost from the study, five for being below the WISC-III critena and two
because English was their second language (ESL). Within the NC Group, ffteen of the
45 shidents were removed. Eight students were eliminated for high WISC-III scores
(seven on Block Design, one on Vocabulary), three students were eliminated for low
Block Design scores, one for low reading, two for ESL and one for physical reasons.
Thus, in the end there were 30 students in both the NC and MD Groups. Having 30
students in the MD Group constitutes a prevalence rate of 16%for this sample. This rate
is somewhat higher than the level reported in studies of elementary sutdents but it may
reflect the increased level of math difficukies in the intermediate and secondary grades
other researchers have noted (e.g., McLeod & Armstrong, 1982). It is also possible that a
higher proportion of poor math students agreed to take part in the study in order to take
advantage of the remediation offered or that these students had poor math teaching in the
elernentary grades, so that the higher number of students with N'RAT3 Anthmetic
standard scores of 89 or below reflects a "teaching disability".

Table 2

Gmup Requirements
Group 1 'WRAT3 WSC-III

Anthmetic Reading Spelling Vocabuiary Block Design FSIQ (est.)

' NC 100- 115 90-110 98-110 8 - 13 8 - 13 280

MD 189 Any score 8 - 13 on either subtest 280


f
A 589 292 292 8 - 13 on eitfier subtest 281)
I I
ARS 1 4 9 1 4 9 1 d39 1 8 - 13 on either subtest
1 280

Stage 2 individual testing continued for the 60 students who met the selection
criteria. The remainder of the test battery t w k approximately 60 minutes to administer,
so that the total Stage 2 testing session lasted about 8û minutes. The test battery is
presented in Table 3.

Measmes
Ail of the measures used in this study were selected as they were believed to be
related to the computation process. Therefore, deficits in any of these measures, relative
to the NC group, may provide some insight into why students with math disabilities have
Table 3

Test Batterg
/ Rocessing
WISC-III :
Coding

WSC-III :
Speed

Symbol Search '

Woodcock Diagnostic
Reading Battery :
Visual Matching
Cognitive Assessrnent
System :
Planned Connections

1
I
1

1
Memou
Children's Memory Scde :
Sequences '

1 Children's Memory Scale :


Numbers "3

Reading Single-Digit
Numbers

Reading Two-Digit
Numbers
.
difficulties in computation. The pattern of strengths and weaknesses among the MD,A
and ARS Groups could be used to develop remediai programs specific to their needs.

I
Mathematics

Differential Abilities Scale :


Sequentiai and Quantitative
Reasonino
KeyMath-R :
Numeration

Estimation

Place Value

Number Facts

' These two subtests scores combine to form the WISC-ILI Rocessing Speed Factor score.
This subtest also leads to scores for Numbers Forward and Numbers Backward.
'These two subtest scores combine to fomi the CMS AttentiodConcentratioa composite
score.

lntellectual Potential Mesure


A
1
1
1

The Vocabulary and Block Design subtests of the Wechsler Intelligence Scale for
Children - Third Edition (WISC-III) (Wechsler, 1991) were administered in order to
detennine an estimated Full Scale iQ score and to have measures of verbal and visual-
spatial abilities for the MD subtyping cornparison.

1) WISC-III :Vocabulary subtest - The Vocabulary subtest contains 30 words of


increasing difficulty that the subject has to define orally.
2) WISC-III : Block Design subtest - The Block Design subtest involves creating
two-dimensional designs with blocks based on a picture model. This is a timed subtest,
with bonus points awarded based on the completion time. This subtest is a measure of
visual-spatial abilities. Rourke (1993) and his colleagues have proposed that Group A
students display deficits in the visual-spatial area which lead to procedural enors such as
poor alignment of numbea.

Written Lan mare Academic Measures


The Wide Range Achievement Test is used in subtyping research in the area of
mathematics. Therefore, the Reading and Speiling subtests of the Wide Range
Achievement Test - Third Edition (WRAT') (Blue Form) (Wdkinson, 1993) were
included in this study in order to determine subtypes of the MD Group based on
academic profiles in reading, spelling and arithmetic.

3) WRAT': Reading subtest - The WRAT3 Reading subtest is a measure of


deooding skills. The subject is required to read aloud single words that are presented all
together on a card.

4) WRAT3: Spelling subtest - In the WRAT3 Spelling subtest subjects are


required to write the correct spelling of dictated words that are presented alone and in a
sentence.

Math Measures
In order to assess math computation, basic math concepts and number fact skills
in grade 7 and 8 students, a variety of math rneasures were inchded in the study. The
WRAT3 Anthmetic subtest pennitted an assessrnent of computation abilities and was
used to identify which students had diffculties in this area. The Sequential and
Quantitative Reasoning subtest frorn the Differential Abilities Scale (DAS) (Bliott, 1990)
and the Numeration subtest from the KeyMath - Revised, Canadian Edition, Form A
(Connolly, 1991) were included as measures of basic math concepts. Two non-
standardized measures of Estimation and Place Value as well as investigator designed
number fact tests for d l four operations were also administered.

5)WRAT3 :Arithmetic subtests - For this study only the written part of the
Arithmetic subtest was needed. It includes 39 math computation questions involving ail
four operations (addition, subtraction, multiplication and division) with whole numbers.
fractions, decimals, and percent as well as algebra and functions. The different types of
questions are intermixeci.
The WRAT3 Anthmetic subtest was used as an independent rneasure to identify
memben of the NC and MD Groups. However, in order to determine which aspects of
math computation led to the lower scores in the MD Group, further investigation of the
Arithmetic subtest was camied out. Therefore, the Arithmetic subtest items were grouped
according to operation and procedure as follows: whole number addition, whole number
subtraction, whole number multiplication, whole number division, fractions,decimals,
percent, dgebra/functions/conversions. An analysis of the types of errors that were made
was also done to determine if the patterns betweeo the groups were similar. Finally a
cornparison of grade 7 and 8 snidents by group was carrieci out in order to see what type
of growth occua over these grades.

6) DAS: Sequential and Quantitative Reasoning subtest - The Sequential and


Quantitative Reasoning subtest measures the participants' ability to determine numericd
patterns. The student is shown two pairs of numbers and must use the sequentiai pattern
in these pairs (e.g., the second number in the pair is one larger than the first number) to
determine the missing nurnber in a third pair. The patterns involve addition. subtraction.
multiplication, division and a combination of these operations. A score is determined
based on the number of correct responses. Further investigation of this subtest was
carried out in order to determine if the operation had an effect on the results. Items were
combined based on the operation needed to complete the sequence forming three groups:
additiodsubtraction, multiplication/division and combination of operations.
7) KeyMath-R: Numeration subtest - The Numeration subtest covers math
concepts such as place value, base-ten, rounding, cornparhg numbers, sequencing
numbers and narning multi-digit numbers. The more difficult questions include
exponents and positive and negative integers. Many of these skilis are thought to be
precursors to the development of computation skills. The items were also clustered into
three categories: number sequences, place value and cornplex numeration (e-g.,
exponents) in order to investigate which basic math concepts differentiated between the
groups.

8) Estimation (see Appendix A) - This test included the two estimation items from
the KeyMath-R which refer to estimation in addition a d subtraction as weU as three
expenmenter created estimation questions for multiplication and division. For each
question, the students' estimations had to be within a certain range in order to be correct.
The ranges from the KeyMath-R were used for those two questions and ranges were set
for the other three questions based on the idea of first rounding the operands to the
nearest tens, twenty-five, hundred or thousand and then estimating the answer. Using
data from al1 of the subjects, this non-standardized measure was found to correlate well
wiîh all the standardized math rneasures included in this study,but particulariy with the
Numeration subtest, with which it has the most in comrnon. See Table 4 for the
correlations.

9) Place Value (see Appendix A) - This is an investigator created test in which the
subjects had to identify the digit that was in the hundreds and thousands place in a
number, round a nurnber to the nearest hundredth and determine the number of tens in a
number wntten in expanded form. This informal measure dso correlateci well with the
other math rneasures and best with the Numeration subtest, suggesting it is measuring a
basic numeration skill. See Table 4 for the correlations.
Table 4
Correlations for Non-Standardaedand Standaroized Math Measare
WRAT3 DAS Key Math-R Estimation Place Value 1
Arithmetic Numeration
I l
p < -001
Place Vaiue .5S .54 .72 -56
pc.001 p < .O01 p < .O01 p c .O01
These corrdations should be interpreted somewhat cautiously as they are based on the
scores of allm students, which represents two discrete subgroups (NC and MD Groups).

10)Number Fact Tests (see Appendix A)


Two different nurnber fact tests were designed by the investigator for use in this study.
In both cases each operation was presented on a separate page.
a) The first number fact test was group administered during Stage 1. Students
were given 60 seconds to complete as many as possible of the 28 number facts presented
for one operation. The procedure was repeated for each operation. The questions
involved the number facts from 2 to 9, including ties (e.g., 4 + 4) and one version of each
non-tie number fact (e.g., only 7 + 6 and not 6 + 7). Those who finished early recorded
their completion t h e . The tests were scored for Mastery Rate (number of questions
answered correctly in 60 seconds). A cornparison between grade 7 and 8 students was
also carried out.
b) During Stage 2, students were a v e n number fact questions in order to
detemine the strategy they used to solve each question and how accurate it was. Ten
horizontal questions were presented for each operation, involving the 6,7,8and 9
number facts. Al1 four ties and one version of each of the 6 non-ties were used. The
questions were listed in random order, wiîh the restriction that no consecutive trials
repeated numencal values in the same position. Students had approximately two seconds
to write down the answer to each question after which t h e they were told to move to the
oext question. After al1 ten questions had k e n attempted, the students went back to those
questions they could not answer automatically and solved them using the back-up
strategy of their choice. The subjects then explained the back-up strategy they had used,
which was recorded by the invesîigator. A number of studies have shown that students
can accurately describe the strategy they have used if they are asked immediately after a
problem has been solved (Siegler, 1987,1989; Siegler & Shrager. 1%). Each question
was scored for accuracy and type of strategy used (e-g., automatic if the student
responded within 2 seconds, slow automatic if the response took longer but was recalled
without the use of a back-up strategy, counting or reference) as welr as the procedure
used (e.g., counting all, counting on, counting-min for addition, counting up or down for
subtraction). Cornparisons were made between the two groups on the following
variables:
Quantity of number facts answered automatically
Patterns of automaticity across operations and fact families
Back up strategies
Errors

Merno- Measvres
A number of different memory measures from the Children's Memory Scale
(CMS) (Cohen, 1997) were included to assess different forms of memory ranging from
rote to sequential and working mernory. Two informal measures of item identification,
which are thought to be related to short-term memory (STM) were also included.

11) CMS: Sequences Subtests - The Sequences subtest measures how quickly and
accurately participants can say both verbal and numerical sequences. It includes verbal
items (e-g., saying the &ys of the week forward and backward) and numerical items (e.g.,
counting from 1 to 10 forward and backward). There is also an item that involves saying
the alphabet and counting in combination (e.g., A 1, B2, C . , etc.). The total score is
based on number of errors and response time for all twelve items together. However, the
individual items appear to be tapping different things ranging from rote and working
memory to sequencing and counting. Therefore, a further exploratory investigation of
this subtest was done to provide some insight into where the differences between the
groups may have occurred. The items were clustered as follows: Forward Number
Sequences, Forward Word Sequences,Backward Number Sequences, Backward Word
Sequences, Counting Sequences and Number-Letter Sequence. The latîer is a verbai task
similar to the visual number-letter items of the Planned Connections subtest. Very little
math research includes sequential tasks, but it seems intuitively logical that sequenrial
mernory would be important for nurnber patterns and recalling procedurd steps. The
Sequences subtest is part of the CMS Attention/Concentration composite.

12) CMS: Numbers Subtests - In the Numbers subtest, participants were asked to
repeat digit sequences of graduated lengths both forward and backwani. The Numbers
Fonvard range in size from 2 to 9 digits, whereas the Numbers Backward range in size
from 2 to 8 digits. There are two trials for each digit length. Scaled scores c m be
calcdated for Numbers Foward, Numbers Backward and the combination of the two,
which is called Numbers Total. It is this ability to detennine scaled scores for both parts
of the subtest that led to its selection over the WISC-III Digit Span subtest. Numbers
Forward is thought to measure short-terni rote verbal rnemory, whereas Numbers
Backward is thought to assess verbal working memory. The Numbers Total and
Sequences subtest scores combine to form the to the CMS Attention/Coocentration
composite.

13)Rapid Number Naming (see Appendix A) - This investigator designed task


had two parts to it. In the f m t part, students had to read nine single-digit numbers (O to
9),presented in random order, as quickly and correctiy as they could. In the second part,
students had to read a randorn selection of two-digit numbers as quickly and correctly as
they could. The times taken to read each set of numbers were used for the analyses. This
non-standardized measure was included as there has been some suggestion that fast
access to information in memory influences STM capacity, with faster item identification
enabling more items to be retained in STM. Both single- and two-digit number tasks
were used because nurnber fact questions involve both types of numbers.
Processine Speed Measures
Processing speed measures were included in this shidy as processing speed is
thought by many to be a factor in both number fact acquisition and computation. The
four processing speed tests that were included in this study, the Coding and Symbol
Search subtests from the WISC-III, the Planned Connections subtest h m the Cognitive
Assessment System (CAS) (Naglieri & Das, 1997) and the Visual Matching subtest from
the Woodcock Diagnostic Reading Battery (WDRB) (Woodcock, 1997), each contain
slightly different factors. Some including numbers, others oniy symbols and some have
more working memory demands. By includmg this variety, the influence of the content
of processing speed measures could be assessed.

14) WISC-III :Coding subtest - The Coding subtest is a timed task which requires
the subject to follow a mode1 and copy geometric symbols that are paired with numbers
into the empty boxes provided. Along with being a measure of processing speed, this is a
paired associate task. Number facts are also thought to be paired associates as the
operands must be associateci with an answer.

15) WISC-III : Symbol Search subtest - In the Syrnbol Search subtest, another
timed task,subjects were required to determine if one of two symbols was present in a
group of five symbols. This is the only one of the processing speed measure that does not
include numbers.

16) WDRB: Visual Matching subtest - This is a visual processing speed test that
measures the participants ability to locate the two numbers that are identicai in a group of
six nurnbers. The numbers increase in length from single-digit b three-digit as the task
proceeds. The score is determioed by the number of correct matches within the three
minute time bit. Sequencing errors and self corrections were dso noted by the
investigator. This measure is unique as it only includes numbers.

17) CAS : PLanned Connections subtest - In the Planned Connections subtest,


students were asked to connect a series of randomly placed boxes on a page. For three of
the test items, the boxes contained numbers that had to be comected in numerical order.
For two of the test items the students had to altemate between boxes containing numbers
and letters, while maintaining the numerical and alpiiabetical order (e.g., 1A, 2B, 3C,
etc.). If a student did not connect the boxes in the correct order, they were toId they had
made an mor and asked to continue from the last correct box. There is a maximum time
limit for each item and timing continued during error instructions. The Planned
Connections score is based on the sum of the time in seconds to cornpiete al1 five items.
This subtest includes processing speed, visual-spatial, working memory and attention
components. Planneci Co~ectionsis similar to the Trail Making Test, parts A (number
only) and B (number-letter) (Reitan, 1966). Harnadek and Rourke (1994) used the Trail
Making Test, Part B as a rneasuE of visual-spatial organization and found thai it was an
area of weakness for Group A individuals. Therefore, separate total time scores for the
numbers-only and number-letter aspects of the Planned Connections subtest were
calculated in order to determine if Group A would demonstrate a deficit in the latter.
CHAPTER 3

RESULTS AM) DISCUSSION

This chapter is divided into rhree sections. In the first section, the demographic
information for al1 groups and an outline of the statistical procedures are presented. ln
the second section, cornparisons between the Nomal Control (NC) and Math Disability
(MD) Groups on the dependent measures are made. In the third section. two MD
subtypes, determined from the WRAT3 academic profiles, are cornpared to each other
and to the NC Group. One MD subtype is cmmposed of sîudents with arithmetic
difficulties exclusively (Group A) and the other subtype includes students with
difficulties in arithmetic, reading and spelling (Group ARS). The comparison of the NC,
A and ARS Groups was a more refined version of the NC versus MD comparison. The
statistics for this thesis were carried out using both SPSS 6.1.1 and Minitab Release
lOXtra for the Power Macintosh.

Section 1

Demogniphic Informafion
The characteristics of the students in the NC,MD, A and ARS Groups are
presented in Table 5. No si+cant sex differences were noted on any of the
dernographic or dependent variables, thus both males and femafes were included in the
grollps.

Com~aiisonsBetween the NC and MD Groups


There were no signifiicant differences between the MD and NC Groups in terms of
age ( F [l, 591 = .W,p > .IO), sex (*[1, N = 60) = 0.267, p z JO), grade (2[l, N = 601 =
0.693,p > -10) or school($ (1, N = 601 = 2.443, p > -10).
Table 5
Chmcteristies of the Four Croups in Stage 2 of the Study
Group NC MD A ARS

Nurnber of 30 30 Il Il
Subjects

Age (years)

Sex
13.2

14/16
13.1

16114
133
5/6
l 12.9

5/6
(Malelfemale)

Grade 7 / 19/11 2Z8 96 1011


Grade 8

School 1 / 10/20 16/14 7/4 7/4


School2

WRAT3 107.0 83.5 85.0 83.1


An thmetic GU (5.4) (3-9) (5.5)

WRAT3 98.4
Reading 1 (5-7)
91.1
(7.5)
98.0
(5.7)
83.4
(3.1)

100.7 88.7 101.2 793


Spelling (5.8) (1 1.8) (6-1) (5.4)

WSC-III 9.9 83 8.3 8.3


Vocabulary 1-4) ( 1-91 (2.4) (1.31

WISC-III Block 113 9.9 9.8 10.2


Design ( 1-41 (2.2) (2.6) ( 1-71

WISC-III 103.5 94.8 94.6 95.4


Estimated FSIQ (6-6) (8-6) (9.6) (6.9)

The selection criteria had significantiy differentiated the NC and MD Groups in


terms of WRAT' Anthmetic skills, with the former scoring signifïcantly higher than the
latter ( F [1,59] = 307.29, p c . m l ) . However, the two groups also differed significantly
on the WRAT3 Reading and Spelling subtests, with the NC Group achieving higher
scores in both cases (F [1,59] = 17.86,p c .O01 and F [1,59] = 25.15, p < -001,
respectively). Therefore, the Reading and Spelling subtest scores were used as covariates
for the anaiyses.
Selection criteria for the hvo WISC-III subtests, which comprise the estimated
FSIQ score, were different for the NC and MD Groups in order to permit MD subtype
cornparisons (which are presented in Section 3). The NC Group had to have scores on
both the Vocabulary and Block Design subtests between 8 and 13, whereas the MD
Group only had to achieve these scores on one of the subtests. As a result, it was not
surprishg to find significant differences between the two groups on both Vocabulary (F
[l, 591 = 13.57,p < . 0 1 ) and Block Design (F [1,59]= 8.57,p < -01).The NC Group
achieved higher s d e d scores in both cases. Differences were also found in the estimated
Full Scale IQ (FSIQ) scores of the two groups (F11,591 = 19.51, p < .001), again in
favour of the NC Group. It should be noted that al1 of the estîmated FSIQ scores were
between 83 and 117. To compensate for the intellectual differences between the two
groups, the estimated FSIQ score was used as a covariate for the analyses.
The covariate analysis was needed to reduce the influence of the group
differences and determine more clearly the effects of disparate math skill levels. In
previous research it has been more common to use only the FSlQ as a covariate, but Bull
and Johnson (1997)found that reading Ievel as a covariate had a powerfut influence and
was aiso by far the most si&~cant predictor of math achievement in their regression
equation. Therefore, reading and spelling were included as covariates dong with the
estimated FSIQ.

cornparisons Between the NC, A and ARS G r o u ~ s


No significant differences were found between the NC,A and ARS Groups on
any of the dernographic features; age (F [2,491= 1.08 ,p z .IO), sex (2[2, N = 521 =
0.01, p > JO}, grade (2[2, N = 521 = 5.15, p > .OS) and s c h o o l ( 2 [2,N = 521 = 4.69, p >
.Os.It should be noted that Group A was composed of alrnost equal numbers of students
from grades 7 and 8, whereas Group ARS was composed mainly of grade 7 snidents. Al1
of the scaled and standard scores used in this study were based on age, not grade,
therefore this grade difference would have Little effect on comparisons using the formai
test measures. However, grade may have had an effect on some of the informa1 math
measures, as the &-de 7 students would have had one less year of instruction. Therefore,
the grade difference should be kept in mind for the informa1 math rneasures when Group
A versus Group ARS comparisons are made.
As would be expected, the mean standard scores for Group A and Group ARS on
the W T 3 Anthmetic subtest were siboniftiwitly lower than the mean standard score for
the NC Group ( F [2,49] = 137.82, p <.01). Significant differences were also found
between the three Groups on the Reading ( F [2,49] = 34.%, p < -001)
and Spelling (F
[2,49] = 59.79,p c -001)subtests of the WRAT3. As hoped, the NC and A Groups were
not ~ i ~ c a n tdifferent
ly from each other and were better than the ARS Group on both
of these subtests. For comparisons between the NC, A and ARS Groups, the Reading and
Spelling subtest scores were not used as covariates as they contributed to the subtype
identification and thus their effect was important.
Both the A and ARS Groups had mean scaled scores on the WISC-III Vocabulary
subtest that were significantiy below that of the NC Group (F [2,49] = 6.28,p c .OS).
This same pattern held for the estimated FSIQ score (F [2,49] = 8.67, p < .ml).
Although the difference between the three Groups on the WISC-III Block Design subtest
was significant (F [2,49] = 3.58,p < .05), none of the posthoc contrasts were signif~cant.
The estimated FSIQ score was used as a covariate in the analyses.

Statisticai Information
Norrnalitv and Homo~eneityof Variance
Before any group comparisons could be made, preliminary analyses of the data
were carried out to determine if there were any violations of nomiality or homogeneity of
variance. Tabachnick and Fidell's (1989)conventional alpha Ievel of .O01 for smail to
moderate sarnple sizes was selected for determining if nomiality was present for each of
the measures in each of the groups. Examination of the distribution of each of the
cognitive and math variables, using the Shapiro-Wilks' Test, indicated that some
measures were not norrnally distributed. Departures from normality, were noted for the
Place Value Test in the NC, MD, A and ARS Groups, Estimation Test in the NC, MD
and ARS Groups, Numben Forward subtest in the NC and A Groups, Reading Two-Digit
Nurnbea in the NC and MD Groups, Numeration and number fact Automaticity and
Mastery Rate for dl four operations but subtraction in the NC Group. The iack of nomal
distribution in the NC Group for many of the math measures was due to the fact that
many of these students scored near the ceiling on these tests. However, the peaked nature
of the results d s o indicated that the mean score was a good representation of the sample
Homogeneity of variance was investigated using the Levene Test. For the NC
venus MD contrast, violations of homogeneity of variance were noted for number fact
automaticity in multiplication and division as weli as overail (Total Automaticity) and for
number fact Mastery Rate in addition. This reflected the srnail standard deviation in the
NC Group, due to the ceiling effect, compared to the MD Group. As the sample sizes of
the NC and MD Groups were comparable, the violations of homogeneity slightly
increased the likelihwd that a siemcant difference would be found between the two
groups on these variables when none in fact existed. These same violations to
homogeneity of variance were also present in the cornparison between the NC, A and
ARS Groups.
A number of diferent transformation were performed on each of the variables
mentioned above (e.g., square root, log, Ikquare mot) but these did not nonnalize the
data. Therefote, the untransformed data were used for the analyses. ALthough al1 of the
measures were left in the study, signif~cantfindings were interpreted with caution as
needed,

A number of test cornparisons were made in this study, therefore omnibus


analyses were performed first. If the muftivariate analyses were significant at the .O5
Level then further univariate and post hoc analyses (Tukey HSD) were c-ed out, with
the sarne p < -05 significance level permitted. However, a more conservative statistical
process which takes into account the number of analyses being carried out would include
using Bonferroni's Correction to determine the significance level of the ANCOVA's.
Using Reading, Spelling and estimated FSIQ scores as covariates, a MANCOVA
was performed for the 24 measures of number facts, math concepts, memory and
processing speed used in the NC versus MD comparison. It was found to be signif~cant
(F [î4,32] = 2.7'76,p < .01)thus ANCOVA's on the individual measures were camed
out. The Bonferroni criterion for these analyses was detennined to be equal to .02. A
second MANCOVA, with FSIQ as a covariate, was done for the contrast between the
NC,A and ARS Groups and it too was signifiant (F [48,52] = 3.149, p c .ml). As the
same number of test measures were used in this comparison, the Bonferroni critenon of p
< ,002 remained. Whenever individual test m u r e s were expiored further, a
MANCOVA was done and a Bonferroni criterion was established in each case.

Section 2
Comp8FISOm Between the NC and MD Groaps

The NC and MD groups were compared in order to determine how students with
math cornputation disabilities differ from students with average math skills. The £ive
main questions used to guide the oomparisons between these two groups are presented at
the beginning of each of the areas.

(i) Analyses of Computation skiils


do grade 7and 8 MD students have trouble with
W h t aspects of cornpuratratron
compared to NC students und do they display 0rdhzr-yor sua su al enor pattern?

In this study, the WRAT3 Arithmetic subtest scores identified students as


memben of the NC or MD Groups. In order to define the cornputation differences
between the NC and MD Groups more specifïcally, an examination of the WRAT3
Arithmetic protocols was carrïed out. This involved a comparison of the types of
questions each group completed successfully as well as the types of emn that were
made.
Corn~utationSkills
The items from the Arithrnetic subtest were divided into eight categories
according to operation (whole number addition, subtraction, multiplication and division)
and topic (fractions, decimals, percent, algebralfunctions/conversions). Cornparisons
between the two groups on these 8 areas were then carried out. It should be noted that
there were only two questions in the decimal and percent categories thus those findings
should be treated cautiously. A graph of the percent of correct, incorrect and not
attempted questions for these 8 areas can be found in Figure 2. An o v e d l MANCOVA
for the 8 areas of the Anthmetic subtest, with reading, spelling and estimated FSIQ as
covariates, was significant (F [8,48] = 14-72,p c .ml).
As repeated analyses were being
done, Bonfer~oni'sCorrection was also computed for these comparisons, resulting in an
alpha level of .06. The rnean scores and standard deviations for the NC and MD
Groups, as well as the statistical results for these 8 computation areas are presented in
Table 6.
No significant difference was found between the NC and MD Groups on the
addition questions (F [1,53= .47. p > .IO). This finding suggests that by grade 7 and 8,
the majority of students had mastered basic addition procedures, including canying.
The NC Group could answer signif~cantlymore subtraction questions than the
MD Group at the p c .O5 level (F[1,551 =739,p = .ûû9), but the difference betw een the
two groups did not meet the more stringent p < .O06 Bonferroni criterion. The
subtraction e m r s of the MD Gmup tended to include sign emrs, number fact e m r s and
mistakes related to regrouping.
A ~ i g n ~ c adifference
nt between the NC and MD Groups was found for both
multiplication (F 11,551 = 34.43, p < .ûû4) and division (F El, 5 3 = 14.68, p c -006).
The NC Group answered significantly more questions correctly in both multiplication
and division than the MD Group. In multiplication the MD Group made number fact,
carrying and procedural errors. (e.g., multiplying in columns; not including a O place
holder in the second row of multi-digit questions). They had particular difficulty with the
2 by 3-digit multiplication question. In division, the MD Group's trouble amse with the
single-digit into multi-digit question and was also evident on a multi-digit divisor
question. They also omitted many division questions. Thus for both of these operations,
the MD Group seemed able to complete the simple questions as well as the NC Group,
but made errors on the more difficult procedures.
Altbough there were some significant differences in whole number computation
skills between the two groups, it was the fraction questions that really seemed to divide
the NC from the MD Group (FII, 551 = 3938, p c .(ME). Between one half and three
quarters of the students in the NC Group could solve addition, subtraction and
multiplication questions involving simple and mixed fractions. In contrast, less than one
third of the MD Group could solve any of these fraction questions. with none being able
to solve the more cornplex questions. The most common error occurred in the addition
and subtraction fraction questions, where the students incorrectiy carried out the
operation on the denominator (e.g., 2/3+ 2 3 = 46).
The questions involving decimals also differentiated between the NC and MD
Groups (F [l, 5n = 17.86,p < -006). These two questions uivolved multiplication with
decimals, thus the MD Group was hindered somewhat by the multiplication aspect of this
task as well as the inclusion of decimals.
The percent questions differentiated berneen the NC and MD Groups but only at
the more liberal p < -05 level (F [l, 5 q = 5.25, p .c -05).The MD Group could barely
answer any of these questions. The algebra, functions and conversion questions involved
skills that are generally not taught until high school, therefore it was not surprising that
both groups found these questions difficult and no significant difference was found (F [1,
5 3 = 1.17, p > JO).
These results suggest that the fraction questions, dong with the more complex
multiplication and division questions seern to be where the grade 7 and 8 students with
math computation disabilities have particular difficulty. These fmdings are similar to
those observed by Ackerman and Dykrnan (1995) in a group of 12 to 16 year olds. It
appears that the MD Group has truuble with the same types of computation questions that
al1 students find challenging (regrouping, complex multiplication and division, fractions),
but it takes them Ionger to master them. The scores and areas of difficulty of the MD
Group indicated that they were performing at a grade 3 to 5 level in anthmetic
computation. Thus although the MD Group appears to have mastered some of the basic
computation skills, they continue to lag behind their peers.
Figure 2
A Cornparison of WRAT3 Arithmetic Questions by Gmup

NC MD MD
Addition Division

.
Fractions
Operations by Gmup

Cf Correct ElIncorrect Not Attempted


Table 6

WRAT3 Arithmetic Item Anaiysis by Operation and Topic Acmss Gronps


OperatiodTOPIC 1 Number 1 NC 1 MD 1 ANCOVA
of Mean Mean (df =1,55)
Questions (Sm (SD) F P
Addition
Computations

.m*
i
Subtraction 5 4.70 3.93
Cornputations (0.6W (0-94.)
r
739
Mu1tiplication 5 4.67 3 .O3 33.43 .001**
Cornputations (0.61) (0.96)

Division 4 3.10 1.83 14.68 .001**


Cornputations (0.61 ) (0.83
I I I 1 1

Fractions 9 433 1.23 3938 .Wl**


( 1-58) (0.9)

Decimals 2 12 3 O.17 17.86 .OOP


(0.73 (038)

Percent 2 I 0.73 0.20 5.25 .026*

Alge braJFunctiens/ 7 O.17 0.00 1.17 11s


Conversion (0-38) (0.W

MANCOVA F [8,48] = 14.72, p c -001


* p < .OS
** p c .006 the significance level set with Bonferroni's Correction for this andysis

Analvsis of Cornputation Errors


The preceding computation analyses indicated that the MD Group continued to
experience somt difficulties with whole number computations. In order to determine
where the computation breakdowns occurred and whetber they were similar across the
two groups, an error analysis was carried out on the twenty questions involving whole
numbers and the four basic operations. Given the number of questions and errors,
Roberts' (1968) and Englehardt's (1977) more general models were used to guide the
e m r classification as opposed to Spiers' (1987) more detailed system. The errors were
classified into six categones: sign,calculation, regrouping, procedure. incomplete and
unknown. Some examples of errors are provided in Figure 3. More than one type of
error could occur on one question. Although most studies only provide the total number
of e m r s made in each category,if one student makes a given error repeatedly the
findings cari be distorted. Therefore, both the number of students making each type of
error as well as the total number of each type of emor were calculated. Chi square
analyses were c h e d out using the number of students in each group who made a given
type of error. The resutts of the error analyses are presented in Table 7.
Sign errors were identifid when a student perfomed the computation using a
sign other than the one presented. The findings indicated that an equivalent number of
students in the NC and MD Groups made these types of erron (2[l,N = 601 = 0.B.
p > .IO). In the majority of cases, the operation the student seiected was the same as the
preceding question (63% of al1 the sign errors). When this did not occur, the sign erron
followed the expected pattern of add-subtract and add-multiply substitutions. This
category also includes Rourke's (1993) Failure to Shift Set category, in which the sign
error matches the sign of the preceding two questions. In the WRAT3 Arithmetic
protocol there are only 3 occasions in which the sarne operation is used for two questions
in a row (-H-; XX-; -x). Sign errors on the first two of these groupings were made by a
comparable number of students in the NC and MD Groups (20%and 17%,respectively).
The calculation category includes both number fact and counting errorç. The
number fact e m r s tended to occur in dl operations but addition. In contrat, counting
errors tended to occur in addition, either in the multi-number addition or multi-digit
multiplication questions. A chi square analysis indicated that signifcantly more students
in the MD Group than the NC Group made at lest one calculation error (2[ 1, N = 601
=4.34,p < -05).
Whenever an error related to carrying or borrowing occurred, they were classified
as regrouping errors. These errors could occur in al1 four operations. This category
included errors such as carrying unnecessarily ,not counting the carry in the caîcdation,
borrowing incorrectly (e.g., borrowing frorn the I O ' S column for the one's column,
usuafly when there was a O in the ten's column) and not writing the cary or crossing out
the number in borrowing and thus seerning to forget it had occurred. Both groups made
these types of errors. There was no significant difference in the number of students in the
NC and MD Groups making these erron (2[ 1, N = 601 = 0.69,p > -10). This finding
suggests that by grades 7 and 8, students with math cornputation disabilities make similar
types and nurnbers of regrouping errors as their p e r s .
E m s such as omitting O place holders in multi-digit multiplication or long
division, multiplying only in colurnns or subtracting the srnailer number from the larger
number regardless of position were classified as procedure errors. A chi square analysis
indicated that more studenfs in the MD Group made procedurai errors than in the NC
Group (2[l, N = 601 =7.50,
p c .01). Al1 of these errors are related to the basic math
concept of place vaiue.
Incomplete questions were those in which students began the question, but did not
complete it (e.g., only determinhg the first number in a long division question). There
was no apparent difference between the two groups on these types of erron (x2[l, N =
601 = 037,p > .IO). The unknown errors category was used for dl of those errors that
could not be classifted into a category as it was not possible to determine how the student
arrived at the answer. Although it is iikely that these errors were due to faulty
algorithms, they were unusual enough that they could not be identified. This category
also included a few cases in which the exact category could not be deterrnined (e.g., 2 x 3
= 5 could be either a number fact or sign error). Due to the small nurnber of students
making unknown errors and its catch-all nature, no chi square analysis was performed for
this category.
The findings from the computation error analyses indicated that the MD Group
generally made similar types of e n o n as the NC Group, but more students h m the MD
Group made procedure and calculation errors. When these results are taken together with
the outcorne frorn the 8 areas of computation, it appears that grade 7 and 8 students with
math computation disabilities are on the same developmental path as their peers, but not
a s far dong its course. The number fact and basic math concept skills of the two groups
are explored further in the next two main comparisons.
Figure 3
Examples of Computation Errors

Incorrect Generaiization No Unneeded Combination Sign


of Procedure CW carr~ Error (x, t),Procedure Error

Number Fact O Emor Procedure Procedure Procedure


Error Error (borrow Error Errors
across O) (big - small)

Procedure Procedure Procedure


Error Error (PV, Error (PV,O as Procedure Error and evidence
Ali gnment) place holder) of Estimation Difficulties
Table 7

The Percent of Stodents in Eaeh Croup Committing Each Type of Arithmetic E m r


Type of E m r NC MD Chi Square P
(df = 1)

Sign 30.0 33.3 0.08 ns


(13) (18)

Cdculation 30.0 56.7 4.34 .O4


( 14) (26)

Regrouping 26.7 36.7 0.69 ns


(8) (12)

Procedure 16.7 50.0 7.50 .O00


(8) (21)

Iacornplete 26.7 20.0 0.37 11s


(8) (8)
1
Unknown 33 233
(1) (7
Total number of errors are in parentheses

(ii) Analyses of Basic Math Concepts


Do MD siudems unders~andbusic math concepts as weil ar NC sr~denfs?

Measures of basic math concepts are rarely included in studies of snidents with
math disabilities. However, Geary, Bow-Thomas and Yao (1992) found that counting
knowledge was related to addition strategies in grade I students. Russell and Ginsberg
(1984) also noted some deficits in base-ten concepts involving larger nurnbers or
estimation in third and fourth graders. Concepts such as place value are the foundation
for more complex math skills, just as letter patterns and phonetic knowledge are the
fourdation for mading words. It is possible that some students do not intuitively master
and generaIize these basic math concepts, just as some students do not see letter patterns
and phonetic rules without having them specifically taught. Therefore, perhaps a lack of
understanding of the basic math concepts contributed to the math difficulties expetienced
by students in the MD Group. These basic math concepts are thougbt to contribute
particularly to the procedural aspect of computation. but may also affect the selection of
back-up strategies for the retrieval of number fa&.
The results indicated chat the MD Group had lower mean scores than the NC
Group on almost all of the math concept rneasures. Significant differences were found
for the KeyMath-R: Numeration subtest (F [l, 5 3 = 21.52, p c .002),the DAS:
Sequential and Quantitative Reasoning subtest (F [l, 5 3 = 15.53, p i-002) and the
investigator created Estimation test (F [l, 5q =14.84, p c .002), but not for the
investigator developed Place Value rneasure (F [l, 551 = 2.52, p > .IO). In the case of the
latter, it appears that with ody 4 questions, one of w hich involved decimals and was
difficult for most students, this test may not have had a good enough range to
differentiate between the two groups. However, the Numeration subtest also included
questions related to place value. The mean scores, standard deviations and results of the
univariate analyses for the basic math concepts are displayed in Table 8.
On the Estimation test, the MD Group had few strategies to cope with the
estimation questions related to multiplication or division, usually resorting to trying
p e n d and paper techniques in their head or guessing. These same strategies were also
noted in poor estimators in previous studies (Rubinstein, 1985; Thredgill-Sower, 1984).
This finding suggests that the MD students would be Less able to use estimation to help
them narrow the possible answers in long division questions. The MD Group's WRAT3
Arithmetic protocols included multiplication sums in the blank spaces mat appeared to be
related to attempts to determine the answers to long division questions (a sample of this is
included in Figure 3). It is also possible that dong with lack of knowledge, the extra
effort needed to compensate for poor estimation skills led to the omission of many
division question.
Table 8

Cornparison of NC and MD Gmups on Measmes of Basic Math Concepts


Math Measures
/ NC(SD)
Mean 1 MI)
Mcan (SD) ,
F
ANCOVA
(df= 1,255)
P
1

DAS - Sequential & 58.27 45.90 15.53 .OOl**


Quantitative Reasoning (7.48) 6-06)
.Cm**
Key Math -
I I
7.77
Numeration ( 1.72)
t 1
Estimation 3.O0 1-13 14.84 .001**
(Max = 5) (1.02) (1 -07)

Place Value 2.97 1.97 2.52 11s


(Max = 4) (-93)

MANCOVA :F [24,32] = 2.776,p < .O1


* p c .O5
** p c . 0 2 , the signif~cancelevel set with Bonferroni's Correction for this andysis

To fuaher investigate the group differences on the Numeration subtest, the


questions on this test were divided into three categories based on what they appeared to
be measuring: number sequencas (e-g., 410,42û, ?, 4-40),place value (e.g., counting
objects displayed in groups of hundreùs, tens and ones) and complex numeration (e-g.,
exponents). The overall MANCOVA was si&cant ( F 13,531 = 5.949, p < . 0 1 ) and
the Bonferroni criterion was set at p < .017.The MD Group scored ~ i ~ c a n t iower
ly
on the place value cluster (F [l,551 = 13.66,p < .01) and the number sequences grouping
(F [1,55]= 4.01, p <.O%, although the latter did not meet the more conservaiive
Bonferroni criterion. No differences were found on the complex questions as both groups
found these difficult (F [ l , 551 = 1.67,p > .IO). The results for the ihree Numeration
item categories are presented in Table 9. Therefore, it appears that compared to the NC
Group, the grade 7 and 8 MD students' difficulties lay more with the place value
questions than the number sequences questions. The latter are usually taught earlier in
the curriculum and are more related to counting patterns.
Table 9

Cornparison of the NC and MD Groaps on the Numeration Sobtest Item Clusîem


Group NC MD F P
Mean (SD) Mean (SD)

Number Sequences 3.90 3.43 4.01 .OS*


(4 questions) (0-30 (0.57

Place Value 6.23 4.13 13.66 .O0P*


(7 questions) (-78) ( 1 .65)

Cornplex 1.O0 0.50 1-67 ns


(htegers and Exponents) (0.59) (0.57)
(2 questions) I
MANCOVA :F f3,53}= 5.949, p < .O01
* p<.M
** p c ,017,the signifieance level set with Bonferroni's Correction for this analysis

One of the items thai the MD Group seemed to have pacticular trouble with on the
Numeration subtest and on the non-standardized Place Value Test was related to the
decomposition of a number. The number had to be represented in expanded form, but the
students had to compensate for a deficit in the hundreds column (e.g., 576 = 4 hundreds,
-tens, 6 ones) or a surplus in the tens column. O d y three students (10%)in the MD
Group h e w how to answer this type of question, although most could expand a number
in the standard way (e-g., 576 = 5 hundreds, 7 tens, 6 ones). In contrast, 20 students
(67%)in the NC could answer these questions. This appears to be an example of
incomplete understanding of the base-ten system leading to an inability to geaeralize the
procedure to similar questions.
An item andysis of the Sequeutid and Quantitative Reasoning subtest was also
carried out to determine if the group differences applied to al1 operations. The questions
involving positive whole numbers were divided into three categories based on whether
the relationship between the numbers involved addition or subtraction (+/-),
multiplication or division (xk), or a combination of operatioos. Once again the overall
MANCOVA was significant ( F [3,53]= 6.626,p c .ûûl) and the Bonferroni criterion
was set at p c -017.The results indicated that the MD Group figured out fewer of the
number patterns in al1 three categories, but only in the combination category did the
difference reach the Bonferroni critenon (+/- : F [L, 551 = 4.82, p < .OS; XI+ : F [l, 551 =
5.98,p c -05;combination : F [1.53 = 12.91, p < .O
1). The mean scores, standard
deviations and statistical results for the Sequential and Quantitative Reasoning subtest
categories are presented in Table 10. Observation of the students during administration
of this subtest suggested that the NC Group seerned to "see" the patterns much faster thaa
the MD Group, thus it would have been interesting to have collected response times.

Table 10

Cornparison of the Croups on the Sequential and Quantitative Reasoning Items


Group
I NC I M D i F I p
+/-
( 10 questions)
1

d+
(5 questions)
4.43
(0.68)
2.47
( 1-08)
I 5.98 .018*

I
I

Combination 1.83 O. 10 12.91 .ml**


(5 questions) (1.37) (0.31)
I 1 1 I

MANCOVA : F 13,535= 6.626,p c .O0 1


* p < .O5
** p < -017,the significance level set with Bonfernoni's Correction for this analysis

The current findings indicate îhat in general, the MD Group had significantly
lower scores than the NC Group on measures of basic math concepts. They did not
answer as rnany estimation, place value, number pattern or number sequence questions
correctly. An incomplete grasp of place value could lead to procedural errors such as
those described in the WRAT3 Arithmetic error analysis ( c g . , not using regrouping when
it is called for and not including a O place holder in multiplication or division). Although
these procedures can be memonzed. previous research suggests that an understanding of
the concepts behind the procedures seems to reduce algorithm errors and enable students
to generaiize the çtrategies to other similar questions (e.g., Hiebert & Weame, 1996).
Estimation skills can be beneficid for determining if answers are reasonable and for long
division. Number sequences and patterns are often the basis of early number fact
procedures. With the building block nature of rnathematics, delays in the mastery of
basic math concepts can negatively affect the acquisition of the next level of skills.

(fi) Anaiyses of Nnmber Fact Data


Are the MD Group's numbm fact speed, atttomaîicity, back-up strategies and
errors similm to those of the NC Group and are t h q consistent across operations?

Previous research has documented the difficulties students with math disabilities
experience with nurnber facts (e.g., Ashcraft, Yamashita & Aram, 1992; Goldman et ai.,
1988). In this study, calculation errors, which included number fact errors, was one of
the two types of cornputation errors made more frequently by the MD than the NC
Group. Therefore, further investigation of the nurnber k t skills of these two groups was
carried out in alL four operations. It is believed that this is the first study to look at dl of
the operations at once. A number of different aspects of number fact knowledge were
investigated, such as speed, automaticity, back-up strategies and errors, in order to
determine how the NC and MD Groups differed and if the patterns were similar across
operations.

Tirned Number Fact Mastery


In Stage 1 of the study, the students were given 60 seconds to complete as many
number facts in one operation as they could. The number fact Mastery Rate of each
student was determined based on the number of correct questions cornpleted in 60
seconds. A Group x Grade comparison was carried out for each of the operations in
order to determine if the NC Group's Mastery Rate was significantly higher than that of
the MD Group, if the grade 8 students had a higher mean Mastery Rate thm the grade 7
students (although it was recognized tbat different students represented the two grade
levels) and if there was an interaction effect. The analyses yielded significant effects for
Group (F 14,501 = 9.53,p i.O0l), but not for Grade (F 14,501 = 9.53,p -c -001) or Group
by Grade interaction ( F [4,501= 9.53,p < .ûûl). Univariate analyses of covariance
indicated that the NC Group significantiy outperfonned the MD Group in subtraction (F
[1,551= 20.90, p < .OM), multiplication (F [1,53 = 25.42, p < .CU?and
) division (F 11,
53 = 18.57,p = .002),but not in addition (F [l, 5 3 = 3.43, p > -10). The Mastery Rates
for each group in each operation are displayed in Figure 4. The lack of grade effect was
likely due to the fact that limiteci gains were made within the NC Group because grade 7
students were already near the ceiling for Mastery Rate.
In order to explore the growth that took place over this one year span more
closely, univariate cornparisons between the grade 7 and 8 students were made within
each of the groups. No signifïcant differences were found between the grade 7 and 8 NC
students, supporting the viewpoint that their lack of improvement contributed to the lack
of overall grade effect (addition F [1,28] = 1.62,p w .IO; subtmction F [1,281 = 4.0 1, p
> .OS; multiplication F 11,281 = 0 . 0 3 , >
~ .10; division F [l,281 = 0 . 1 8 , ~
>.IO). A
univariate cornparison of the grade 7 and 8 students within the MD Group indicated that
the grade 8 MD students completed significantly more number facts successfully in 60
seconds than the grade 7 MD students in ail operations but subtraction (addition F 11,281
= 4.84, p i.05; subtraction F [l,281 = 234, p > .IO; multiplication F [l,281 = 6.42, p c
.05;division F [l, 281 = 7.47, p < .OS). However, none of these results met the more
conservative Bonferroni Criterion of p < .O1 set for these analyses. Despite these gains,
the lack of Group x Grade interaction effect suggests that the gap between the NC and
MD Groups was not significantly reduced. The Group by Grade results are displayed in
figure 5.
These findings suggest that the NC and MD Groups cao answer a similar number
of questions in 60 seconds in addition, but not in the other operatiom. Even though some
differences in number fact speed were noted between the grade 7 and 8 MD students,
they are not becoming simcantly closer to the number fact speed of the NC Group in
subtraction, multiplication and division. In order to compare how the NC and MD Group
determine their answers to number fact questions, an investigation of number fact
strategies was carried out using the Stage 2 Number Fact Tests.
Figure 4
Cornparison of Number Fact Mastery Rate
Between the NC and MD Groops

Addition Subtractiiin Mu1tiplication Division


Operation
Figure 5
Gmop by Grade Cornparison of
Nnmber Fact Masterg Rate

Grade

---O--

I --*--
MD Addition
MD Subtraction
1 - - - A - -MD Multiplication
!
--+-- MD Division
-
-NC
-NC

-NC
Addition
Subtraction
NC Multiplication
Division
2
1

1
Number Fact Automaticity
For each of the number facts given to the students in Stage 2, it was determined,
by observation and asking the student to describe their strategy, whether he/she knew the
fact automatically or if a back-up strategy was used to figure it out. If the student
responded irnmediately (less than approxirnately 2 seconds) the response was designated
as automatic. If the student counted the numbers (e-g., 6 + 9 = 6,7,8,9,
10, 11, 12, 13,
14, 15) it was cailed a counting strategy, which is equivalent to Levels 1 and 2 in
Fuson's (1997) system. If the student referred to a known fact to help determine the
answer, it was labeled as a reference strategy. The latter is considered more sophisticated
than the former and is ranked as Level3 in Fuson's system. A fourth coding label, slow
automatic, was used if it took the student more than 2 seconds to corne up with an
answer, but no specific strategy was reporteci. The percentage of each strategy used for
each operation are presented in Tables 11 to 14.
The mean number of questions answered with automatic reûieval was compared
for the NC and MD Groups and for grade 7 and 8 student. across al1 four operations. The
Group x Grade ANCOVA was significant for Group (F[4, Ml = 10.26, p c .O01), but not
for Grade (F [4,50] = 0.95, p > .IO)or Group by Grade interaction (F [4,5û] = 1.05, p z
.IO). Once again it is likely that the ceiling effect in the NC Group contributed to the lack
of grade effect. Investigation of the grade 7 and 8 differences in automaticity were
therefore carried out within each group. No signifïcant differences between the grade 7
and 8 NC students were found, but the grade 7 NC students were already very close to the
ceiling on this task so there was little room for improvement (addition F[1,28]= 0.65,p
< -10;subtraction F[1,28] = 0.15, p > -10; multiplication F[1,28] =O. 19, p z -10;
division F[1,28] = 0.62, p > .IO). Cornparison of the grade 7 and grade 8 MD students
indicated that the latter used automatic recall for significantly more number facts than the
former in al1 operations, but o d y division met the Bonferroni Criterion of p c .O1
(addition F [1,28) = 4.30. p < -05; subtraction F [ l , 281 = 4.55, p < .OS; multiplication F
[ l , 281 =4.82, p < .OS; division F [1,28] = 9.45. p -c .(Il). However, the lack of
significant Group by Grade effects suggests that these gains were not large enough to
significantly reduce the difierence between the NC and MD Groups. The Group by
Grade results are displayed in Figure 6.
The Group effect indicated that the NC Group used automatic retrieval for
significantly more number facts than the MD Group. Univariate analyses demonstrated
that this was true for dl four operations (addition F [l,551 = 23.75,p < .ûû2; subtraction
F [ 1,551= 12.89; p 4.002, multiplication F [1,53 = 34.84, p < .002: division F [ 1,53
= 24.31. p < -002).Based on previous research (e.g., Reischner, Garnett, & Shepherd,
1982) this fmding was not unexpected. The number fact automaticity levels of the two
groups are displayed in Figure 7. The NC Group relied on automatic recall for alrnost al1
of the 6 through 9 tables, whereas the MD Group knew only 40 to 60 percent of the
number facts automatically.
A separate analysis cornparing the autornaticity levef of the different operations
within each group was signifïcat in both the NC ( F [3,27= 6.44, p c.002)and MD
Groups (F [ 3 , 2 7 = 9.64, p < .001). Wlthin the NC Group, an inspection of the mean
scores suggested that they knew fewer facts automatically in subtraction than the other
three operations. Whereas in the MD Group, the operation effect seemed to stem from
the higher level of automaticity in addition compared to the other three operations. It is
likely that this higher level of addition automaticity, dong with more efficient use of
back-up strategies, contributed to the lack of difference between the two groups for the
addition Mastery Rate. Addition is usually the first operation that is taught. Therefore,
the increased exposure to the addition facts rnay have contributed to the higher level of
automaticity in this operation. If thk is m e , then the use of automatic retrieval in the
other operations may also improve. With number fact speed important for success in
higher level math (Atkinson, 1983; Geaiy & Wideman, 1987) these gains may irnprove
the chances of success for the MD Group.
These results suggest that the NC Group used automatic retrieval for significantly
more number facts than the MD Group. Even though the grade 8 MD students knew
more number facts than the grade 7 MD students, the gap between the NC and MD
Groups remained. The higher automaticity Ievel for addition in the MD Group suggests
that increased exposure may be important, but the improvements seem to be slow and
there may be a Leveling off effect once the ties and easier number facts are Learned. Thus
it is possible the Geary's (1994) proposa1 that number fact automaticity is a deficit, and
not a delay, in MD students is correct.
Table 11
Percent of Each Strategy Used for Addition Nurnber Facts
Group Auto Slow Auto Reference Count Omit

NC 923 0.7 5.0 2.0 0.0

Table 12
Percent of Each Strstegy Used For Subtraction Nomber Facts
Group Auto SIow Auto Reference Count Omit
NC 80.0 3.7 83 8.0 0.0

MD 45.0 03 10.0 44.7 0.0

Table 13
Percent of Each Strategy Used for Moltipikation Nnmber Facts
Group Auto Slow Auto Reference Count Ornit

NC 94.7 2.0 33 0.0 0.0

Table 14
Pement of Each Strategy Used for Division Number Facts
Gmup Auto Slow Auto Reference Count Omit

NC 95.0 2-0 3 .O 0.0 0.0


1

MD 44.7 4.0 21.0 10.7 19.7


Figure 6
Groap by Grade Cornparison of
Number Fact Automaticity

8
Grade
- - * -MD
--*--
MD
--+--
- Addition
MD Subtraction
Multiplication
1 - - + - - MD Division
-
-NC
+NC

-NC
Addition
Subtraction
NC Multiplication
Division
Figure 7
Cornparison of N d e r Fact Automaticity Between
the NC and MD Groups

Addition Subhaction Mu1tiplication Division


Operation
Level of Number Fact Automaticity for Each Ouestion
Previous research has suggested that some number fact questions are easier to
leam than others. A comparison of the number fact autornaticity level for each question
in an operation was undertaken in order to determine if the NC and MD Groups had
similar patterns. The results are displayed in Figure 8 by fact family (e.g., & 6, 12 - 6,6
x 6 and 36 + 6 are d l represented in the 6,6 fact family column). With a fkw exceptions
in multiplication and division, the tie questions were easier for both the NC and MD
Groups to recall automatically. This finding was to be expected based on previous
research (e.g., Ashcraft & Baüaglia, 1978;Graham & Campbell, 1992). The problem
size effect, which suggests that questions with larger numbers are more difficult than
questions with smaller numbers, was less evident wiîh these nurnber facts. Perhaps this
was due to the fact that al1 of the numben included here are considered larger and the
problem size effect is usuaily noted when the smaller nurnber facts (e.g., 2 x 3) are also
inchded.
As is evident in the Figure 8 graph, although the NC Group knew more oumber
facts automatically, the pattern of automaticity was quite similar for both groups and
across al1 four operations. A Spearman Correlation indicated that the NC and MD
Groups had similar rankings for addition (r = .94, p < .ûûl), multiplication ( r = .63, p <
.05)and division (r = .67,p < -05) but not subtraction (r = 5 3 , p > -10).A comparison of
the rankings across operations in each group was also canied out using the Spearmm
Correlation. The rankings of al1 of the operations correlated signifcantly with each other
in the MD Group (r = .68 to .go). In the NC Group, al1 of the between operation
correlations were signifcant (r = .69 to .85) except for the subtraction - division one (r =
35). The exact correlations for both groups can be found in Appenciïx B. This finding
suggests that the two groups generally found the same questions easy to retrieve
automatically and that the pattern was sirnilar across operations. The implications of this
finding for the notion of a common retrieval system will be addresseci in the General
Discussion chapter.
Figure 8
Aritomatization of Number Facts
by Number Fact Families

-
Number Fact Families

--*--
MD Addition -NC Addition
- -a- - MD Subtraction NC Subtraction
- - + - -MD Mu1tiplication * NC Multiplication
l
--*--
MD Division NC Division
Back UV Stratepies
A description of the back up strategies used by the NC and MD Groups is
provided in order to provide a clearer picture of the MD Groups' approach to number
facts and to dernonstrate the similarities between the two groups. More detailed charts
related to the number fact data are presented in Appendix B.

Use of Each Type of Strate~y


As expected based on previous studies (e.g., Bisanz & LeFevre, 1990; Geary &
Burhgham-Dubree, 1989; Goldman, Mertz & Pellegrino, 1989; Lemaire & Siegler,
1995),almost al1 of the students in both the NC and MD Groups used a combination of
strategies to solve the number fact questions for each operation. Only five of the thirty
students in the NC Group used automatic retrievai for al1 of the number facts in ail four
operations; none of the MD students did. Within the MD Group, counting was the most
frequently used back up strategy for addition and subtraction. Al1 but four of the students
in the MD Group used a reference strategy at least once, most often for the multiplication
and division number facts. Therefore, despite the fact that the majonty of the MD
students knew how to use reference strategies, which are considered to be more
sophisticated, counting strategies continued to be used, patticulariy on the two operations
that are taught first. This observation suggests that the grade 7 and 8 MD students seem
to select a strategy that is most efficient for them and stick with it. It appears that Fuson
et al's (1997) mode1 of strategy development for number facts may need some
modification. These will be reviewed in the General Discussion chapter.

Countin~StrateGes in Addition and Subtraction


Counting strategies in addition and subtraction were looked at to see if there were
any differences between the two groups. Al1 of the students used counting on from the
larger number for addition. For subtraction, cornparisons were made between the number
of students using counting up versus counting down. Subtraction is usually taught using
the laîter (e-g., 7 take away 2 = 7,6,5)and students switch to the former when they
determine it is more efficient for questions in which the difference is smaller than the
second number (e.g., 7 - 5 = 2). With the questions asked in this study, the effectiveness
of one counting strategy over the other was minimal. No significant differences were
found between the NC and MD Groups in tems of the proportion of students using
counting up or down ($(1, N = 2001 = 2.91, p > -05).See Table 15 for data on the
subtraction counting strategies. Therefore, the MD and NC Groups use similar
approaches with counting strategies in subtraction. It was dso interesting to note that in
both groups, almost every student used either counting up or down but not both.

Table 15

Nnmber of Sîudento Per Groop IJsing Comting Stnitegies in Sobtraetion


Group Number of Count Up Count Mixture of Total
Students Down Both
NC 30 5 3 1 9
( 15) (5) (4) (24)
MD 30 12 11 1 24
(69) (59) (6) ( 134)
Number of questions answered using a giveo strategy is in parentheses

Reference Stratepjes Across Operations


Reference strategies were also analyzed to determine which number facts were
used as reference points. The results for each operation are found in Tables 16 to 19.
The tie number face seemed to be one of the most commonly used reference facts in
every operation. This is iogical, given that ties were the easiest number facts to recdl
automatically in dmost ail cases. Number facts involving 5 or 10 were aiso used
frequently in every operation, most often for questions involving 9.
It was interesting to see that in subtraction and division students tended to refer to
oumber face in the reciprocal operations (addition and muitiplication, respectively).
When the students were asked about the strategy they had used for the division number
facts, al1 but one of the student (from the NC Group) indicated that they had used
multiplication to determine dl of their answers (e.g., for 42 + 6 they had thought 6 x ? =
42), whether automatic or back-up strategies were used. This observation may shed some
light on the nature of tie effects. In early research it was thought that ties were easier to
leam than non-ties as the repeated operands drew from the same list of answers so there
would be less interference. However, when tie effects were found in division (Campbell,
1997),this theory was catled into question as division questions do not involve repeated
operands. The present observations suggest that grade 7 and 8 students actually refer to
multiplication when they answer division questions, so that the original theory could still
apply. Overall, the reference strategy patterns of NC and MD Groups were found to be
very similar between groups and across operations.

Examples of Number Fact Back-Up Strategies and Errors


Using Ties or Number Facts as a Reference Points

7~9=?,7~6=42,42+6=48,48+6=54(incorrect)
7 x g = ? , g x 2 = l 8 , l 8 + 1 8 = 3 6 , 3 6 + 3 6 = 7 2 , 7 2 + 9 = 8 1(incorrect)

Using 5 or 10 as a Reference Point


8 + 6 = ?, 10 - 8 = 2.6 - 2 = 4, so 8 + 6 = 14J("whatever this [the first value] is away
from 10 you subtract from this one [the second nurnber] and you put a 1 in front of it and
you've got your answer")
6+9=?,"took1frorn6so I O + S = 15"4
15-7=?,8+8=l 6 , l 6 - I = l5s07+8= 15,therefore 15-7=8J

Tricks for the 9 Times Table


Each finger is assigned a number
from 1 to 10, beginning with the baby
finger of the left hand. The finger that
represents the number being multiplied by 9 is bent-dowi. The number of fingers to the
Left of the bent finger represents the tens digit. The number of fingers to the nght of the
bent finger represents the ones di@t(see diaO- above).
The tens digit of the answer is one less than the number being multiplied by 9. The
sum of the tens and ones digits of the answer equals 9, so the ones digit = 9 - tens digit.
Table 16

Nomber of students using Each Type of Reference Strategy for Addition


Ties 9 Trick

l (23 I ( 16) I (3 I
Number of questions answered using a given strategy is in parentheses

Table 17

Nomber of students esing Each Type of Referwce Stmtegy for Snbtraction


Group Ties 5 or 10 Number facts 9 Trick
+ - + - + -
NC 4 1 I 4 3 O O
(7) (1) (1) (8) (6)
MD 8 O 1 2 4
1 O
(13) (2) (6) (5) (1) .

Number of questions answered using a given strategy is in parentheses

Table 18

Number of students asing Each Type of Reference Strategy for Mdüplication

l Tics l 5 or 10
l Numberfacts
9 Trick

I (18) (22) ( 17) l (9)


Number of questions answered using a given strategy is in parentheses

Table 19

Nomber of students using Each Type of Reference Strategy for Division

l Tics i 5 or 10
i Numberf=
9 Trick

1 (23) (12) (20) (8)


Number of questions answered using a given strategy is in parentheses
Number Fact Error Analvses
The specific findings from the emor analyses are available in Appendix B, but a
brief discussion will be presented here. In subtraction, no difference in the number of
studeots making errors was found, regardless of whether counting up or counting down
strategies were used, but it should be noted that more counting down errors were made
overail. En multiplication it was expected that table-related errors would be the most
cornmon type of error (Campbell & Graham, 1985). This was found to be the case for
both the NC and MD Groups. Aithough both groups made table and close errors, oniy
the MD Group made errors rhai could not be classified.
Previous research had suggested that number fact errors tended to be one operand
away from the correct answer in addition, multiplication and division (e-g., 6 + 8 = 13; 6
x 8 = 42; 48 i 6 = 7) (e-g., Campbell, 1997; Geary, 1990; Siegler & Robinson, 1982).
The findings from this study demonstrated that this pattern held for al1 four operations in
both groups, except subtraction in the NC Group. In subtraction, the NC Group tended to
make errors when the question included a 9 (e.g., 16 - 9), seeming to treat the 9 as an 11
(e-g., 16 - 9 = 5). This resulted in a higher level of answers that were hvo operands away
from the correct answer. Generally, for addition and subtraction, the answers were
under- or over-counted by one approximately haif of the tirne. For multiplication and
division the answers were one number fact away from the correct answer at least three
quarters of the time.
In a cornparison of the proportion of number fact errors made by students in each
group (number of errorshumber of questions answered), the mean score of the NC Group
was lower than that of the MD Group (4.67% and 12.73%, respectively), but no
significant difference was found (F [ 1,551 = 130, p > .IO), in part due to the large
standard deviations (4.09 and 11-63,respectively). It was noted, however, that a large
percentage of the number fact errors occurred on non-ties for both the NC and MD
Groups (82%and 72%, respectively).
A nurnber of researchers have suggested that MD students make more errors
when they rely on automatic retrieval than NC students (e-g., Geary, 1996; Fleischner,
Garnett & Shepherd, lm).An overall MANCOVA indicated that there was a
significant difference between the NC and MD Groups in ternis of the percent of
automatic erron made (F [4,46]= 2.636, p < -05).The Bonferroni criterion for the four
cornparisons was set at p c .O 1. Univariaie analyses indicated that there were no
differences in the error rates for addition (F [ l , 491 = -04,p > .IO), subtraction (F [l, 491
= 3.16,p > .05)or division (F [ 1,491 = 3,
p > -10). A s i ~ c a ndifference
t was found
for multiplication (F [1,49] = 6.33, p c -05)but it did not meet the more stringent
Bonferroni alpha level. Thus in general, although the MD Group seemed to make a
higher percentage of errors in automatic recall, with the exception of multiplication, the
difference in each operation was not significant. Ermr rate cornparisons for the other
number fact retrieval strategies were not camied out as the NC Group used back-up
strategies so rarely.
The overall findings from the number fact analyses indicate that the MD Group
continues to lag behind the NC Group in the use of automatic retrieval for number facts
in d l four operations. The MD Group also answered fewer number fact questions in 60
seconds than the NC Group in al1 operations but addition. There are indications that the
MD Group is still making gains in both number fact automaticity and speed, but despite
this they continue to remain well below the level of the NC Group. The pattern of
autornaticity across d l four operations was found to be sirnilar for both groups, with ties
generally easier to retneve automatically. The types of counting and reference back-up
strategies used by both groups were comparable. The similarities between the two groups
in ternis of tie effects, strategy use and types of enors suggest that the MD Group is
following the same developmental path as the NC Group. However, whether they will
continue to make gains in number fact automaticity that will enable them to catch up
remains to be seen and may depend on cognitive factors such as memory and processiog
speed.
(iv) Analyses of Cognitive Variables
Are there differences between the NC and MD Croups in the cognitive areas of
memory and processing speed?

Memory
A number of researchers have proposed that memory deficits, particularly in
working memory, are associated with number fact autornaticity and computation
procedures (e.g., Geary, 1990;Geary & Brown, 1991; Siegler & Shrager, 1984).
However, the results of previous studies on dl types of memory in students with math
disabilities have been Mxed (Bull & Johnson, 1997: Fletcher, 1985;Siegel & Ryan,
1989). Few studies have included sequential memory tasks. However, this type of
memory w o d d seem to be relevant to math in terms of recalling procedural steps in order
and Ieaming number sequences.
Despite the trend for the MD Group to obtain Iower mean scores than the NC
Group, significant differences were found on o d y one subtest of the Children's Memory
Scale, the Sequences subtest (F [l,5% = 11.78, p < .002). This contributeci to a
si,@icant CMS Attentionlconcentration composite score (F [l, 551 = 5.01, p ç .OS). No
signifcant Mferences were obtained for Nurnbers Fonvard (F(1,551= -16,
p >. IO),
Backward(F [l, 5Sj =.03, p > -10) orTotal(F [l, 551 = .06,p > -10). The twoinformal
masures, Reading Single-Digit and Two-Digit Numbers did not differentiate between
the NA and MD Groups either (F 11,551 = 3.47, p > .O5 and F [1,551 = 2.%, p > .OS,
respectively). The memory findings are presented in Table 20.
Few studies of math disabilities have included sequential tasks. In this study, the
Sequences subtest seemed to distinguish between the NC and MD Groups very well, with
the former scoring significantly higher than the latter. This subtest includes 12 items
which seemed to tap rote sequential mernory, working memory and counting memory.
Thus a closer inspection of the Sequences subtest was caried out in order to determine
the source of the differences between the groups.
Table 20

Cornparison of NC and MD Gmnps on Memory Tests


Memory and Attention NC MD ANCOVA
Tests Mean (SD) Mean (SD) (df = 1 , s )
F P
CMS - 107.27 93.43 5.0 1 .029*
AttentionlConcentration (9.26) ( 11.46)

CMS - 9.80 8.27 0.06 ns


Numbers Tota1 (2.22) (2.64)

CMS - 10.17 897 O. 16 ns


Numbers Forward (2.02) (3 -20)

CMS - 9.80 8.27 0.03 ns


Numbers Backward (2.60) (3-00)
CMS- 12.60 9.60 1 1.78 .Wl**
Sequences ( 1 -92) (2.w
Read Single Digit 2.98 3 -81 3-47 11s
Numbers ( T h e in sec.) (0.63) (1.14)

Read Double Digit 5.08 6.54 2.96 ns


Numbers ( T h e in sec.) (1.18) (2.55)

MANCOVA : F [24,32] = 2.776,p < .O1


* p < .O5
** p < ,002, the sigoificance level set with Bonfernoni's Correction

Results of the Sequences item analyses are presented in Table 21. Regardless of
whether numbers or words were involved, no signifcant difference between the NC and
MD Groups were found on forward sequences (F [l. 551 = 0.42, p > -10for numbers and
F [l, 551 = 0.10, p z .10 for words) or backward sequences (F [l, 5 3 = 0.07,p z .10 for
numben and F [ l , 551 = 1.75,p > -10 .for words). Nor were signifïcant differences
found for the number-letter combination item (F [ 1,551 = 3.63, p > .Os.These findings
suggested that the differences between the two groups did not lie with the rote or working
mernory sequential tasks. However, signifiant differences were discovered for counting
by a aven nurnber (F [l, 551 = 10.48, p < -003).Thus the NC and MD group differences
on the Sequences subtest seemed to occur primarily on the counting task. This type of
skip counting (e.g., 3 , 6 , 9 , 12,...)would be similar to what is often doue when a counting
strategy is used to determine the answer to a multiplication question. This approach is
often used in the initial stages of multiplication mastery.

Table 21

Comparison Between the NC and MD Groups on Items h m the Seqaences Sobtest


1

NC MD F P
Forward Number 6.93 6.83 0.42 ns
( m a = 7) (0.25) (0.38)
Forward Word 17.60 16.87 0.10 ns
(max = 21) (237) (2.74)
Backward Nurnber 1 1.37 10.67 0.07 11s
(max = 14) (1.10) ( 1.29)

Backward Word 11.9'7 10.37 1.75 ns


(max = 14) (2.13) (2.76)
Counting 15.97 10.17 10.48 .002**
(max = 21) (3-76) (4.13)
Number-Letter 3.O7 1.50 3 -63 11s
Combination (2.52) ( 1.72)
(max = 7) 1
MANCOVA : F [6,5U] = 2.449, p < .O5
* p<.05
** p < .008,the significance level set with Bonferroni's Correction for this analysis

Thus, the current findings indicated that there were no significant differences
between the NC and MD Groups on measures of verbal short-term or working memory.
The lack of working memory differences between the two groups was particularly
surprising, giver. its proposed role in number fact acquisition. It has been thought that
deficits in working mernory hinder the development of strong associations between
operands and aoswers thus inhibiting number fact automaticity. This result sheds some
doubt on this proposal. The differences between the two groups on the Sequences subtest
supgests that this may be a factor in cornputations and number fact acquisition.

Processing speed measures were included in the current study as it was believed that
the ability to think and work quickly was one of the cognitive characteristics that
influenced nurnber fact acquisition and computation abilities. Previous research bas
suggested that students with difficuities in math tend to process information more slowly
than their peen (Ackerman & Dykman, 1995; Bull &Johnson, 1997; Gross-Tsur et al.,
1995). However, each of the past studies used different critena to define math disabilities
and different comparison groups. Thus it was hoped that the current shidy could further
inforin the processing speed findings. Research in other areas (e.g., memory) has
indicated that the content and mode of presentation of a test can affect whether
differences between groups are found. Therefore, four different processing speed
measures were Uicluded in this study, each of which brought a distinctive component to
the processing speed task. The Visual Matching subtest was unique as it included only
numbers whereas the Symbol Search subtest was the one measure that had no numbers.
The Coding subtest included numbers in a paired associate task. Finally, the PIanned
Connections subtest included a visual-spatial component as well as working memory
demands for some of the items. This variety should make it possible to detemine if the
content of the test influenced the outcorne.
The results of the ANCOVAs, controlling for estimated FSIQ, Reading and Speiiing,
indicated that the NC Group achieved significantly higher mean scores than the MD
Group (at the p < .O5 level) on aU of the processing speed tests (Coding F [1,55] = 11.37,
p c .002;Symbol Search F [l, 5 3 =5.79, p < -05; Visual Matching F [l, 551 = 16.58, p -=
-002and Planned Connections F [1,55J = 10.21, p < .002),as well as the Processing
Speed Factor from the WISC-III (F[l, 551 = 10.03. p < .002). The mean scores, standard
deviations, and statisticai information for the processing speed measures are presented in
Table 22. This outcome strengthens the current view about processing speed deficits in
individuals with math disabilities and extends this finding to measures not previously
used in MD research (Visual Matching and Planned Co~ections).
If the more conservative p c -ûû2Bonferroni criterion was used however, then the
difference on the Symbol Search subtest was no longer significant. As this is the ody
processing speed measure that does not include numbers, it appears that the presence or
absence of numbers does influence the results. When numbers are present in processing
speed measures, the NC Group scores ~ i ~ c a n thigher
l y than the MD Group, but if
there are no numbers, as in the S ymbol Search subtest, the test does not differentiate
between the groups as well.

Table 22

Cornparison of MD and NC Stodents on Proeessing Speed Tests


Processing Speed NC MD ANCOVA
Tests Mean (SD) Mean (SD) (df = 1,55)
F P
WSC-III - 116.20 97.67 11.37 .001**
Processing Speed ( 12.65) ( 12.09)

WSC-III -
Coding
t

WSC-III - 13.03 9.60 5.79 .019*


Symbol Search (2.50) (2.93)

WDRB - Visual
Matching

CAS - Planned 12.73 9.63 10.42 .002**


Connections (238) (2.28)
- -

MANCOVA :F [24,32] = 2.776, p < -01


* p-C.05
** p c -002, the significance level set for this andysis with Bonfemoni's Correction
The present fmdings indicated that students with math disabilities scored
significantly lower than normal cootrols on tests of processing speed. However, the
inclusion of numben in a processing speed task may further hinder the MD Groups
ability to work quickly. Perhaps it is slower processing speed that prevents the
association between operands and answers in number facts acquisition and the
completion of procedurat steps in computations.
The overail cognitive picture indicates that there are differences between the NC
and MD Groups in the areas of sequential memory and processing speed, but there were
some unexpected outcornes. The NC Group outperformed the MD Group on the
sequential memory test, but the two groups were not significantly different from each
other on the verbal short-term or working rnemory tasks. The lack of difference between
the NC and MD Groups on the latter was especiaily surprising. The NC Group did
significantly better than the MD Group on the mesures of processing speed and it
appears that the inclusion of numbers in the task may also have been important.
In order to make the cornparisons between the NC and MD Group easier to
visuaiize, the test scores were converted into z scores based on the standardized means
and standard deviations for each measure. These are displayed graphically in Figure 9.

(v) Regression Analyses


How much of the WRAT3 Arithmefic score c m be accounted for by the cognitive
and math variubles incltuied in this stuùy?

The findings from the NC versus MD cornparisons indicated that the NC Group
generally outperf'onned the MD Group in basic math concepts, number fact automaticity
and speed, processing speed and sequential memory. Although differences between the
NC and MD Groups were found on these masures, their contribution to the computation
skills of these students was unknown. Therefore, regression analysis was used to explore
the relationship between the cognitive and academic variables in this study and the
WRAT3 Arithmetic score. Correlations (using a two-tailed ~ i ~ c a ntest)
c e between the
math, cognitive and written language variables for the entire sample are presented in
Table 24. These correlations should be interpreted cautiously as they are based on the
combination of two distinct subgroups of students (NC and MD Groups). The first
column contains a11 of the correlations to the Arithmetic subtest and the last row contains
al1 of the correlations to the total number fact automaticity score.
Figure 9
Scores on the Cognitive and Acadcmic Variables for the NC and M D Groups

Test Measures

Ari = Arithmctic, Rd = Reading, Sp = Spelling, V = Vocabulary, BD = Block Design; Cd = Coding, SS = Symbol Search, PC = Planncd
Connections, VM = Visual Matching, S e q = Sequences, NF = Nurnbers Fonvard, NB = Numbers Backward, Num = Numeration, S t Q =
Sequential and Quantitative Reasoning
The correlations between tests, along with the theoreticaï framework behind the
tests, were used to cluster some of the variables together for use in the regression
analysis. This was done as the sample size of 6û limited the number of predictors that
could be used and it was felt that a composite score may be more robust than an
individual subtest score. A11 of the test scores were converted to z scores, based on the
standard deviation and means of this sample, to equalize the influence of each score on
the composite score. A Math Concept Composite score was formed from the
Numeration, Sequential and Quantitative Reasoning, Estimation and Place Value tests.
The four processing speed tests : Coding, Symbol Search, Visual Matching and Planned
Connections were combineà to create a Processing Speed Composite score. The CMS
Attention/Concentration Composite score was used to represent the memory areas. The
WRAT3 Reading and Speiling scores were combined to form a Written Language
Composite score. The number fact automaticity scores from ail four operations were
combined to form a Total Autornaticity score- The estimated FSIQ score was the final
variable included in the regression andysis.
With the exception of the Total Automaticity Composite, the alpha levels from
the Anderson-Darling Normality Test for d l of the composite scores were above the
conventiona1 level of - 0 1 suggested by Tabachnick and Fideil (1989)for smdl to
moderate size samples. A ceiling effect prevented the Total Automaticity Composite
score from meeting this requirement. Inspection of the studentized residuals also
indicated that the requirements for linearity, equality of variance and normality were met.
A step-wise regression was performed to detennine the order in which variables
should be entered into the regression equation. The first regression that was carried out
included none of the math variables, in order to detennine what factors aside from math
were related to cornputation skills. The findings indicated that the Processing Speed and
Attention/Concentration Composites, along with the estimated FSIQ score al1 led to
significant changes in the R2 and resulted in a overall adjusted R
' of 62.4%. Therefore,
these three cognitive variables can explain almost two thirds of the variance in the
WRAT3 Anthmetic computation scores of this group of grade 7 and 8 students. The
results of this first regression are presented in Table 23.
Table 23

Remsion Anaiysis for WRAT3 Arithmetic Withont Math Measnres


Test R2
adj) 1 adj change) change

Processing
Speed (49.8)

Attentiod

Written
Language

The second regression included aii 6 variables as predictors of the WRAT3


Arithmetic scores. The results can be found in Table 25. In this regression, the Total
Number Fact Automaticity, Math Concepts Composite and Processing Speed Composite
scores d contributed significantly to the variance in the computation scores. These three
variables explaineci 76.9% of the variance. It was not surprising to fmd that Total
Number Fact Automaticity was the first predictor as it is highly correlated with the
WRAT3 Anthmetic subtest. However, it is also highly correlated with the basic math
concepts, thus it was surprising to find the latter as the second predictor. This order
suggests that the Basic Math Concept Composite must contribute some unique elements
that are related to computation skills but not aumber fact autornaticity. This idea is
consistent with the view thrit basic math concepts are related to the procedural aspect of
computation. The importance of processing speed in computation is aiso suggested by
the results of this regression analysis as well as the fint analysis. Processing speed could
be a factor in both number fact acquisition and the procedural aspect of math
computation.
Table 24
Correlation Matrix

A R S V BD Cd SS VM PC Sq NF NB 1-D 2-D NU S i Q Est PV


A
R
S
v
BD
Cd
SS
VM
PC
sq
NF
NB
1-D
2-D
Nu
S+Q
Est
PV
Auto

A/C

A = Anthmetic; R = Reading; S = SpeUing; V = Vocabuiary; BD = Blwk Design; Cd = Coding; SS =


Symbol Search; V M = Visual Matching; PC = Planned Connections; Sq = Sequences; NF = Numbers
Foward; NB = Numbers BacLward; 1-D = Singie Digit Reading; 2-D= Two-Digit Reading; Nu =
Numeration; S+Q = Sequential and Quantitative Reasoning; Est = Estimation; PV = Place Value; Auto =
Total Nurnber Fact Automaticity; A/C = AttentionIConcentrationComposite; NT = Numbers Total

Two-tailed correlation
Table 25
Regression Analysis for WRAT3 Arithmetic with ail Measorps
Test R' R' Change F P F P
adj) ad adj change) change change
Total 66.0 10.60 <.OOI 11235 c.001
NumberFact (65.4)
Automaticity
Math 75.0 9.0 4.54 ~.001 85.44 4 01
Concepts (74.1) (8-7)
Composite
Processing 78.1 3.1 2.82 .O07 66.53 <.O0 1
Speed (76.9) (2-8)
Composite
Attention/ 78.6 0.5 1.1 1 2.72 50.4 1 <.O0 1
Concentration (77.0) 1)
(0-
Composite
Wntten 78.6 0.0 035 .725 39.7 1 <.O0 1
Language (76.6) (-0.4)
Composite
FSIQ 78.6 0.0 0.22 .829 32.52 < .O01
(76.2) (-0.4)

Section 3

(vi) Cornparisons Between the NC,A and ARS Croups


Are there subtypes within the MD Group that display different paîterm of
strengths und weaknesses?

The cornparisons between the NC and MD groups on the cognitive and math
measures indicated that the MD Group scored significantly lower than the NC Group in
processing speed. sequencing. numeration, estimation, number patterns and number fact
automaticity and speed. This type of cornparison treats the MD Group as if it were
homogeneous. However, the MD group was composed of students with different
configurations of acadernic strengths and weakness, so that combining them together may
have masked some of the potential differences. Rourke and his colleagues (e.g., Rourke,
1993; Rourke & Finlayson, 1978; Strang & Rourke, 1983),found that students with
disabilities only in math displayed different cognitive deficits than those with disabilities
in math, reading and spelling. Thus. if the MD Group is considered to be a
heterogeneous group composed of unique subtypes, comparisons between the NC Group
and the subtypes should be made. In this study, two MD subtypes had sarnple sizes that
were large enough to pennit useful statisticd analysis. The first subtype consisted of
eleven students with deficits ody in arithmetic (Group A) and the second subtype
included eleven students with deficiis in arithrnetic, reading and spelling (Group ARS).
As a result of their research findings, Rourke and his colleagues have suggested
that Group A students display a deficit in the visual-spatial area, whereas Group ARS
students have difficulties in the verbal area or even development in both areas.
Therefore, the f m t analysis that was done looked at the Block Design and Vocabulary
scores of the two subtypes to see if that pattern ernerged in this study. The NC, A and
ARS Groups were then compared on al1 of the cognitive and math measures used in the
NC versus MD cornparison. An overview of the findings wiil be presented here. with the
results of these comparisons available in Appendix C.

Cornparisons of Vocabulary and Block Design Discre~ancvPatterns


The flexibility of allowing students in the MD Group to demonstrate average
ability on either the Block Design or Vocabulary subtests permitted an investigation into
the profile of the A and ARS subtypes on these two WISC-III subtests. Based on
previous research (e.g., Rourke, 1993; Share, Moffit & Silva, 1988), it was expected that
the Group A students would dernonstrate better Vocabulary than Block Design abilities
( V S D ) , whereas the Group ARS students would show the reverse pattern (BD>V) or no
difference between the two subtests (V=BD). The MD shidents were grouped according
to whether they scored 3 or more points higher on Block Design or Vocabulary or
wbether the point spread was less than 3. A three point diifference between the
Vocabulary and Block Design scaied scores is sibdcant at the .O5level for the 12 to 14
year age -pup (Sattler, 1992). T h e outcome of this breakdown are presented in Table
26.
Table 26

-
Namber of Students in Each of the Vocabuiary Block Design Gmops
V and BD NC MD A ARS
Relationship
V>BD O 4 2 O

V = Vocabulary; BD = Block Design

The findings from the ARS Group were consistent with what was expected, with
students displaying lower scores on Vocabulary (BbV)or no difference between the
two subtests (V=BD). However, it was very surprising to find that Group A was not
composed prirnarily of students with better Vocabulary than Block Design scores
( V S D ) . In fact, only two of the Group A students demonstrated this pattern, the
rernaining members had no difference or a Block Design advantage. The remaining two
VzBD students fit the profile of the AS Group, which was one of the subtypes that was
too srnall to include in subsequent analyses. The cornparison of Vocabulary and Block
Design scores led to one of the interesting findings of this study. The rnakeup of the
Group A students in this study suggested that it is possible for grade 7 and 8 students to
have a specific deficit in arithmetic and not have poor visual-spatial skills.

Results from the NC. A and ARS Group Analvses


C o m ~ s o n between
s the three Groups were made on al1 of the cognitive and
math variables. In the computation analyses, the NC Group outscored the A and ARS
Groups on the multiplication, division, fractions, and decimals questions, whereas no
si,@icant differences were found on the addition or subtraction questions. This result is
similar to the NC versus MD findings, with the exception of the lack of subtraction
diference. This suggests that the students with poor subtraction skills did not rneet either
of the subtyping cntena. The NC Group outperformed the ARS Group. but not the A
Group, in percent, however as the ARS Group was compoçed mainly of grade 7 students
who have had less exposure to percent, this result is not unexpected. Most of the
Anthmetic errors for both the A and ARS Groups were related to calculation or
procedure. Thus in generai, the A and ARS Groups performed similarly to each other on
the WRAT' Arithmetic subtest and in most cases below the level of the NC Group.
The NC Group scored significantly higher than both the A and ARS Groups on al1
of the basic math concept rneasures, including the Place Value Test. No significant
differences were found between the A and ARS Groups on any of the tests, suggesting
that these two subtypes exhibit similar deficits in their understanding of basic math
concepts-
Some new patterns emerged in the timed number fact comparisons. In the Stage 1
number facts, the NC and A Groups had equivdent Mastery Rates in addition and both
were better than the ARS Group. In subtraction and division the NC Group was better
than Group A which in tum was better than Group ARS. In multiplication, the NC Group
couid answer more questions in 60 seconds than both MD subtypes. Therefore, in tenns
of number fact speed, the three groups seem to represent a continuum of abilities with the
NC Group at the top, the ARS Group at the bottom and the A Group in between.
The results of the number fact automaticity comparisons were similar to those in
the NC venus MD contrat. The NC Group answered ~ i g ~ c a n tmore
l y number facts
with automatic retrieval than the A and ARS Groups in al1 four operations and the A and
ARS Groups did not diifer from each other- As there were no differences between the A
and ARS Groups in automaticity. it seems that Group A may use number fact strategies
more efficientiy allowing them to do better than the ARS Group in timed situations. The
small sample sizes prevented a profitable analysis of the number fact error patterns.
The NC Group scored significantly higher thm both the A and ARS Groups on al1
of the measures of processing speed. Once again the Symbol Search subtest was not
significant at the more conservative p < .ûû2 Bonferroni criterion. Although the trend
appeared to be that Group A scored higher than Group ARS on these tests, this difference
was o d y significant on the Visual Matching subtest.
In the area of memory. some different patterns became apparent. Significant
differences between the NC Group and the math subtypes were present ou tbree measures
where no NC Group venus MD Group differences had been found: Numben Fonvard,
Reading Single-Digit Numbers and Reading Two-Digit Numbers, aithough the first two
tests did not meet the Bonferroni criterion of p < .002. In al1 three cases, post hoc
comparisons indicated that the NC Group scored ~ i ~ c a n thigher
l y than the ARS
Group, but not the A Group. On the Nurnbers Fomard and Reading Two-Digit Numben
measures, Group A also sigpif~cantlyoutscored Group ARS. Thus, Group ARS
displayed deficits in rote verbal memory and accessing numben from long-term memory,
whereas Group A performed simiiarly to the NC Group in these measures. Accessing
information from long-term memory has been suggested as a factor in short-terni memory
(STM) performance, as the faster items can be identified, the iarger the number of items
that can be retained in STM (e.g., Bull & Johnson, 1997; Case et al., 1982; Hitch &
McAuley, 1991). The finding of deficits in verbal STM in the ARS Group is consistent
with previous research studies (Fletcher, 1985; Hamadek & Rourke, 1994, Shafrir &
Siegel, 1994). The fact that these deficits were found only in the ARS Group suggests
that they may be more related to reading and spelling than math. The remainder of the
memory findings were similar to îhose noted in the NC versus MD comparisons. The NC
Group scored significantiy higher than both the A and ARS Groups on the Sequences
subtest and no significant differences were found on the Numbers Backward subtest-
The results of these subtype analyses suggested that the A and ARS Groups in this
study were more alike than different on al1 of the cognitive and math measures, with the
exception of the STM related tasks. Thus, the MD group was more homogeneous than
expected. Although rarely statistically s i g d k r & the trend on many of the cognitive
measures was for Group ARS to score lower than Group A. Thus. the ARS Group
seemed to be a weaker version of Group A. This becomes clearer when the data are
presented in a ,gaphical form. Therefore, the scores for each of the three groups were
converted into z scores, based on the standardized means and standard deviations, so that
they could be displayed on the same graph (see Figure IO). This finding would seem to
be more consistent with a normal variation than subtyping model, an idea that will be
elaborated upon in the General Discussion chapter.
Figure 10
Scores on the Cognitive and Academic Variables for the NC,A and ARS Croups

r 1 Ir
I1 Y - - t --- -----
- - 1 --
- T - - - - -7------l --

FSIQ V BD Cd SS PC VM Scq NF NB Num StQ


WISC-III Proccssing Speed Memory Math

Test Measures
1 +NC -a- -A - - - A - -ARS 1
Ari = Arithmetic, Rd = Reading, Sp = Spclling, V
= Vwbulary, BD = Block Design; Cd = C d i n g , SS = Symbol Scnrch, PC = Plnnncd
Connections, VM = Visual Matching, Seq = Scqucnces, NF = Numbers Fonvnrd, NB = Numbcrv Backward, Num = Numcration, S t Q =
Sequcntial and Quantitative Rwsoning
CHAITER 4

GENEML DISCUSSION

The results from this study suggested that grade 7 and 8 students with math
cornputaiion disabilities made signifcantly more calculation and procedural e m on the
WRAT3 Anthmetic subtest than their peers with average math computation skills. They
also had more trouble with the multiplication, division and fraction questions.
Investigation of the factors thought to be related to math computation indicated that the
MD Group achieved simcantly lower scores than the NC Group on meastires of basic
math concepts, number fact speed and automaticity, sequential memory and processing
speed. These same areas of difficulty were also found in both of the MD subtypes
(Groups A and ARS). The basic math concepts, number fact automaticity and processing
speed factors predicted a substantid amount of the variance in the math camputation
scores. The implications of each of these findings are discussed below. A potential
frarnework for math computation based on the results is also presented.

(i) Math Computation


Previous research has suggested that students with math disabilities make slow
progress in the area of mach computatioo (e.g., Cawley et al., 19%; Cawley & Miller,
1989; Ackerman et al., 1977). In the present study, the grade7 and 8 MD Group achieved
grade equivdent scores on the WRAT3 Anthmetic subtest ranging from grade 3 to grade
5, indicating a delay in these students of between two and five years. This result is
similar to that found by Ackerman and Dykman (1995) using the WRAT-R Arithmetic
subtest and only slightly higher than might be expected based on Cawley and Miller's
(1989) findings wiîh the WSPS Calculations subtest. Therefore, it appears that the MD
Group selected for this sîudy was similar to that found in other studies involving
adolescents in t e m of the degree of their computation difficulties.
The first goal of the present study was to determine what aspects of math
computation grade 7 and 8 students with MD have diffculty with compared to the NC
Group. A comparison was made between the number of WRAT3 Arithmetic questions
the two groups answered correctly in each of the four operations as well as for fractions,
decimals, percent and algebra/functions/conversions.The results of this investigation
indicated that the NC Group aoswered sieecantly more questions correctly in whole
number subtraction, mukiplication and division, as well with fractions, decimals and
percent, although the subtraction and percent fuidings did not meet the more conservative
p < -006Bonferroni critenon. This outcome concurs with the finding of multiplication.
division and fractions difficulties in adolescents by Ackerman and Dykman (1995).
Studies involving younger students suggest that MD students have trouble with
regrouping in addition and subtraction (e-g., Russell & Ginsburg, 1984; Saito, 1992). In
the current study, the students with MD were at the same ski11 level as the NC in addition
and approaching them in subtraction. If this finding is combined with those from
previous studies, it appears that the MD Group rnay have a developmental delay in the
mastery of whole number computations so that with tirne they rnay also catch up to the
NC Group in multiplication and division. Fractions include a part-whole component that
the MD students rnay take longer to grasp, just as they took longer to master part-whole
concepts with whole numbers. Perhaps it is the lack of mastery of fractions that prevents
the students with MD from achieving grade equivalent scores beyond the junior grades
even when they are near the end of high school.
The second part of the computation analysis involved comparing the error
patterns of the NC and MD Groups. Siecantly more shidents from the MD Group
made calculation and computation errors than in the NC Group. The high level of
computation errors in the MD Group is consistent with the Russell and Ginsburg's (1984)
results. Like the findings of Cawley et al. (19%), many of the procedure errors noted in
the current study involved O, usuaily when it was k i n g used as a place holder in
multiplication or division. These rnistakes and other procedural erron such as
multiplying oniy in columns seem to be related to base-ten place value concepts. Kouba
et al. (198û) also noted that many procedural errors seemed to be due to a lack of
understanding in these basic math concepts.
Unlike the study by Saito (1992) involving younger children, in the present study
there was no significant difference in the number of students in the two groups making
regrouping errors. As the NC and MD Groups showed no difference in the number of
addition questions answered correctiy and a borderiine difference in subtraction, this
finding was not unexpected. However, given that regrouping has been found to be
troublesome for younger students with MD, the lack of significant difference in these
grade 7 and 8 students tends further support to the notion of a developmental delay in
computation development-
No simcant difference was found in the number of students from each group
making sign errors. In the majonty of cases, the students used the same operation they
had performed on the preceding question. This finding is different than that of Roberts
(19681, who noted that better math students made fewer sign errors. Perhaps the
difference in these results stems from the fact that the students in Roberts' study were
younger than those in the current study and thus had to focus more attention on the basic
math computation to solve them correctly.
It appears that the students in the MD Group tended to make similar errors as the
students in the NC Group, but more students with MD made errors, in particular
calculation and procedural errors. It also seems that the aspects of computation that the
MD Group finds difficult (e-g., regrouping, long division) are the same parts that al1
students find chaüenging, but it takes the MD Group longer to consolidate them. The
findings from the computation analyses seem to suggest that the MD Group is on the
same developmental path as the NC Group, but not as far dong its course.

(fi) Basic Math Concept.


A number of mathematics rneasures thought to assess some of the fundamental
math concepts were included in this study. The NC Group scored significantly higher
than the MD Group on the KeyMath-R Numeration subtest, which includes questions on
number sequences, rounding, place value and applying base-ten d e s to counting. The
poorer understanding of base-ten and place vafue concepts displayed by the MD Group
may account for their larger number of procedural errors. This relationship has been
found in previous studîes (e.g., Carpenter et al., 1997). Hiebert and Weanie (1996) found
that a poor understanding of place value concepts hindered students ability to generalize
and modify strategïes to new questions. Thus perhaps there were some ornitted questions
on the WRAT3 Arithmetic subtest that the MD Group could have done if they had
understood the procedures enough to generalize the strategies. According to Hiebert and
Wearne's (19%) findings, it is also possible that the MD students forgot some of the
procedures as they were relying on memory as opposed to understanding.
On the DAS Sequential and Quantitative Reasoning subtest, the MD Group found
it harder than the NC Group to determine the relationship between pairs of numbers.
This finding suggests that the MD Group had more trouble with number patterns and
determining part-whole relationships. It is possible that because the students with MD
did not know their number facts as well as the NC students, the patterns between groups
of numbers were not as obvious to them.
The MD Group also did significantly worse on the Estimation Test. Unlike the
NC Group, they tned to appfy the procedures they had leamed for paper and pencil tasks
or resorted to guessing, instead of rounding the numbers. This observation is similar to
that noted in previous studies involving estimation (Reys et al., 1982; Rubinstein, 1985;
Threadgill-Sower, 1984). The MD Group's poor estimation skills were also evident on
the WRAT' Arithmetic questions. Many of the students with MD needed to make a few
multiplication atternpts to figure out how many times a divisor went into a dividend as
opposed to rounding the numbers and estimating (e.g., for a question such as 3317 + 64
they would try 64 x 3 = 192, CA x 4 = 256,64 x 5 = 320 to get the f m t part of the answer
instead of thinking 30 + 6 = 5 and using that estimation to guide them).
With signifcant differences between the NC and MD Group at the more
conservative p < .003on the Numeration, Sequential and Quantitative Reasoning and
Estimation measures, it appears that students with MD do not understand basic math
concepts as well as the NC students. Studies involving younger children indicated that
they had trouble with some of the eariy basic concepts such as conservation and couoting
mles. Due to the fact that the basic math concepts tend to build on one another (e.g.,
without an understanding of conservation principles, it is difficult to understand how
numbers can be decomposed, which in tum makes the concept of place value difficult to
epsp), it is possible that students with math disabilities fail behind their pers at the
earliest stages of number sense and then are forever trying to catch up andor build upon a
shaky foundation.

(Ui) Number Facts


With the inclusion of number facts from al1 four operations, cornparisons between
the NC and MD Groups as well as between operations could be made. Of al1 of the
measures given, number fact automaticity was the most highly correlated with the
WRAT' Anthmetic score. Developing number fact automaticity has been identified as
an area of difficulty for students with math disabilities (Ashcraft, Y amashita, & Aram
lm,Geary et al., 1991; Goldman, 1988; Howell et al., 198'7). Therefore, it was not
surprising to fmd that the MD Group knew significantly fewer number facts, from the 6
through 9 fact families, automatically than the NC Group. This was hue for every
operation. According to Siegler's mode1 (1986) the lower level of automaticity in the
MD group suggests that they do not have enough confidence in the accuracy of their
answer to respond automatically andlor that their operand-answer associative strength is
not peaked, perhaps because the operands have been associated with a variety of different
answers due to cdculation enors. In the current data the emor rate of the NC and MD
Groups was not sigdicantIy different, but this was k e l y due to the wide variability
between subjects within each group. Therefore, it is not possible to determine if higher
numbers of errors were a factor in the flatter distribution. It was noted, however, that a
much Iarger proportion of the errors for both groups occurred on non-ties (82% for the
NC Group and 72% for the MD Group) which may account for the lower level of
automatic retrieval on these questions. It is also possible that the lower level of automatic
retrieval for the MD group is due to their slower processing speed so that the operand-
answer associations are weak (by the tirne the answer is determined the operands have
faded from memory).
The findings from this study provide support for both the delay and deficit models
of automatic number fact retrievai. Two results suggest that the MD Group may be
delayed in their developrnent of automatic retnevd. The MD Group knew significantly
more number facts automatically in addition (60%)than the other operations (40%), and
it is addition that is usualIy taught first. Therefore, with extra time, improvements in
automaticity could also be noted for the other operatioos. The second findimg was the
fact that within the MD Group, the grade 8 students generally retrieved more number
facts automatically than the grade 7 students. Akhough these grade findings are based on
cornparisons between different students, they suggest that the potential for continued
growth in number fact automaticity is still present in the intermediate grades.
There are also findings that support the deficit model. Subtraction, multiplication
and division were introduced at different grade levels yet the automatic retrieval level for
these three operations seems to have reached a similar plateau. Therefore,perhaps, the
ievel of automatic retrieval in addition is higher because it is learned first so there is less
proactive interference. Secondly, although there were gains withùi the MD Group, they
were not significant at the more stringent p < -01level, nor were they large enough to
create a sisnificant Group by Grade interaction. Thirdly, the automatic retrieval level for
the non-ties was low across operations for the MD Group, with the exception of the 8.9
fact family in multiplication and division (where they could use the 9 tricks and number
patterns). They also made a high proportion of their errors on non-ties. Therefore, it will
be harder to create a peaked distribution for these number facts and gain enough
confildence in the correctness of the answer to respond with automatic retrieval for non-
ties. It appears that the MD Group learns the tie questions automatically, as they are
easier to leam than non-ties, but masters few of the more dificult facts. This observation
is consistent witb the MD Group's mean automatic retrievd rates of 4 to 6 as four of the
ten number fact questions given involved ties. Finally, the similarities in the back-up
strategies of the two groups suggests that the MD Group is not doing anythmg different
than the NC Group in this regard, yet they have been less able to switch to automatic
retrieval. Perhaps the slower processing speed of the MD Group is preventing this
transition from occuning.
For ai1 but five students in the NC Group, who used automatic recail for every
number fact, a mixture of strategies was used to answer the number fact questions. The
use of a strategy mix by most students was expected based on previous research (Fuson,
1997). When number facts were not known automatically, either counting or reference
back-up strategies were generally used. Within the MD Group, four students only used
counting strategies, placing them at Level2 in Fuson et d.'s (1997)stages of number fact
development. Since these four students also knew some number facts automatically,
Fuson et al.% (1997) view that automaticity could develop at any level appears to be
correct,
The remaining MD students used the more sophisticated reference strategy at
least once, usually to answer the multiplication and division questions. However, despite
knowing this strategy, counting was the most frequently used back-up strategy for
addition and subtraction for al1 but one of these students. Therefore, many of the MD
students appear to be at both Level2 and Level3 depending on the question or operation.
Perhaps students do not necessarily progress from Leve12 to Level3 unless the counting
strategy is inefficient or inaccurate. Counting strategies in multiplication and division
would seem to be more time consuming and error prone as they involve large numbers,
thus for these operations reference strategies would likely be more effective.
Interestingly, the students in the NC Group only used reference back-up strategies for
these two operations. In conhast, counting strategies can be used quite quickly and
accurately in addition d subtraction.
The methods used within the counting and reference back up strategies were
sirnilu in both groups. With addition, al1 of the students knew how to count on from the
larger number. An equal proportion of students in the NC and MD Groups used counting
up or down in subtraction. When the NC and MD students relied on reference strategies,
ties or the numbers 5 or 10 tended to be used as referent facts the most often. As ties
appeared the easiest for dl of the students to memorize, it was not surprising to find them
used as referent facts. This finding was consistent with that noted in Amencan students
by Fuson (1997). The ten facts were used most often when the number 9 was involved in
the question as Steinberg (1985) had also found in a U. S. study. For both subtraction
and division the students tended to refer to facts from the reciprocal operation (addition
and multiplication, respective1y).
Ptevious researchers have hypothesized that ties were easier to learn due to the
fact that the operands are the same and therefore draw from the sarne table, reducing the
interference from other tables. M e n tie effects were discovered in division, where the
operands are not the same, this hypothesis was questioned (Campbell, 1997). However,
in the current study it was observed that alI but one student used multiplication to solve
al1 of the division number fact questions, regardless of the retnevai strategy selected.
Therefore, while the division facts are being automatized it appears that students refer to
multiplication, in which case the repeated operand hypothesis about tie effects is still
feasible.
In a tirned situation, the NC Group completed signifïcantly more number facts
correctly than the MD Group in dl operations but addition. This suggests that the MD
Group seems to have caught up to the NC Group in addition number fact speed, the
operation they have been doing for the longest penod of t h e and in which they have the
highest level of automaticity. It is possible that improved efficiency in the use of back-up
strategies also contributed to this gain in the MD Group. Here too, srnall gains in
Mastery Rate were noted between the grade 7 and 8 MD students, but they were not
substantial enough to create a significant Group by Grade effect.
A cornparison of the number fact rankings across operations was also c d e d o u t
The percentage of students answering each nurnber fact automatically was calculated.
These percentages were then used to determine a ranking for each number fact for each
group, in al1 four operations. Similarities in the rankings were found between the NC and
MD Groups for addition, multiplication and division, but not subtraction. Thus the NC
and MD Gmups generally displayed similar patterns of automaticity. It was found that
number facts belonging to the same fact family (e.g., 6 + 6, 12- 6 , 6 x 6 and 36 t 6)
usually had similar rankings in ail four operations, for both the MD and NC Groups. For
both groups the ties were generally the easiest number facts to r e c d automatically and
the non-ties were harder to memorize. In order to compensate for this, more time couid
be spent learning the difficult facts. There is some debate in the number fact literature
about whether each operation uses an independent retrieval network or whether there is a
more integrated network. Campbell (1997)had proposed that the reciprocal operations of
multiplication and division may share a retrieval network. The high correlations between
corresponding questions in al1 four operation, as well as the similarities in error patterns
(e.g., the high proportion of incorrect answers that are one operand off in al1 four
operations) suggest that there may be a common integrated retrieval network across al1
four operations.
The overall findings from the number fact analyses indicated that the MD Group
used similar back up strategies and made sirnilar types of errors as the NC Group, and
that b t h groups generaliy found tie questions easier to answer than non-ties. However,
despite these similarities in number fact acquisition, the MD Group remaineci behind the
NC Group in number fact automaticity (dloperations) and speed (al1 operations but
addition). Although some evidence for continued improvements in automatic retrieval
were found, they seem to be ovenbadowed by the evidence for a deficit in this area.
Given the association between number fact automaticity and speed with success in
secondary and college mathematics that has been found in other studies (Atkinson, 1983;
Geary & Widaman, 1987) the MD Group appears to be at risk for continued diculties
in mathematics.

(iv) Cognitive Variables


The cognitive variables of processing speed and memory were selected for this
study as deficits in these areas are thought to be present in students with math disabilities.
l y than the MD Group
The results indicated that the NC Group scored ~ i ~ c a n thigher
on measures of processing speed, confirming the findings from the two previous research
studies on processing speed involving slightly different populations. However, when the
more conservative Bonferroni Correction was applied, this outcome did not apply to the
WISC-III Symbol Search subtest, the only processing speed mesure that did not include
numbers. This suggests that it may not simply be that students with math disabilities
process information more slowly, but that they process information involving numbers
more slowly. The influence of test content on outcornes is not new (e.g., previous
memory findings), but this is the first time it has been noted with processing speed
measures in a study of math disabilities.
Rocessing speed may be a factor in number fact acquisition. Geary and Brown
( 1991) found that a group of gifted grade 3 and 4 students executed their counting
strategies faster than the nomial control group. They proposed that this faster counting
speed may aiiow the operand-answer associations to become stronger and thus enable the
shift to automatic retrievd to occur. Therefore, in the current study, the slower
processing speed in the MD Group may have reduced the strength of the operand-answer
associations and thus have hindered their shift to automaticity, resulting in the lower level
of automatic retrieval for al1 operations. If this is the case, then the MD Group's levels of
number fact autornaticity are likely to remain low.
A number of different verbal memory tasks were incIuded in this study. One of
the most surprising findings was the lack of significant differences between the NC and
MD Groups on the CMS Numbers Backward subtest, a measure of working memory.
This same pattern was also evident on the working memory items, involving backward
words or numbers, of tbe CMS Sequerices subtest. Revious research had proposed that
working memory was important in the initial stages of number fact mastery, when backup
strategies were king used (Geary, 1990; Hitch & McAuley, 1991; Siegler & Shrager,
1984) and potentially in computation procedures.
Based on the findings from this study, it seems that it may not always be working
memory that hinders the development of number fact automaticity; in some cases it may
be slow processing speed. Some students with math disabilities may be able to hold as
much information in working memory as NC students, but their slower processing speed,
taxes their working memory, so that in the case of number facts, the operands and
answes do not get as strongly associateci. This rnay be the case for the MD Group in the
current study, as their number fact automaticity level was signif~cantlybelow that of the
NC Group.
Although the findings have k e n inconsistent, short-term memory deficits have
been found in some of the studies of math disabilities, particularly if the tasks included
digits or spatial cornponents (Fletcher, 1985; Siegel & Ryan, 1989). In the current study,
no significant differences in verbal STM were found between the NC and MD Groups on
the CMS Numbers Forward subtest. This is consistent with the STM finding of Bull and
Johnson (199'7) who also covaried out reading in their study. When the MD subtype
cornparisons were made, the ARS Group performed signifïcantly below the level of the
NC and A Groups. Therefore, perhaps some of the inconsistent STM findings in MD
studies have k e n due to the different compositions of the MT3 Groups in each of the
studies. Those with a higher proportion of Group ARS students in their MD Group
would be more likely to find STM differences than those with a higher proportion of
Group A students.
Sequential memory is the final type of memory associated with math disabilities
addressed in this study. It has been proposed as a factor in counting fluency (Geary et al.,
1992; Hitch & McAuley, 1991) and procedure deficits (Geary, 1990, 1993). Although
they did not use formal criteria to identify math disabilities, Bull and Johnson (1997)
found that students with lower math scores did more poorly on alphabet and number
sequences than students with higher math scores. Sequential memory also seemed to be a
factor in the present study as the NC Group scored signifïcantly higher than the MD
Group on the CMS Sequences subtest. It should be noted however, that closer inspection
of the items suggested that much of the p u p differences amse from the counting
questions (e-g., counting by a given number as quickly as possible). Differences were not
significant on the other fonvard and backward word and number sequences, that are more
sirnilar to those used by Bull and Johnson. Therefore, although the overall Sequences
score suggests this is an area of difficulty for the students with math disabilities relative
to the NC group, the types of sequences which are difficult remains somewhat unclear.
Overail, the NC Group scored significantly higher than the MD Group on
masures of processing speed and sequential memory, but no differences were found in
working memory or STM.

(v) Priedictors
For the reg-ression analyses, the tests that measured simikir constmcts were
combined together to f o m six composite scores that were used as predictors. In the first
regression analysis, in which the two math predictors were not included, almost two
thirds of the variance in WRAT3 Anthmetic scores could be explained by the Pmcessing
Speed Composite score, the Attentiodconcentration Composite score from the CMS, and
the estimateci FSIQ score. The hi& predictive power of the Processing Speed Composite
is consistent with the findings of Bull and Johnson (1997). They found that after reading,
only processing speed added significantly to their variance in math scores. Unlike their
findings, the reading/spellinp composite did not emerge as a significant predictor of
computation ability.
When the two math composites of Totai Number Fact Automaticity and Math
Concepts were included in the regression, they dong with the Processing Speed
Composite score explained three quarters of the variance. It is likely that factors such as
teaching, homework completion and attitude towards mathematics contribute toward
some of the remaining variance.

(vi) Math Disabiütg Subtypes


Based on the subtyping Literature, in particular that of Rourke and his colleagues
(e.g., Rourke and Finlayson, 1978),it was anticipated that Group A and Group ARS
would display different patterns of strengths and weaknesses on the cognitive and math
measures used in this study. Group A students were expected to display a visual-spatial
deficit However, only two of the Group A students obtained scores on the WISC-III
Block Design (which measures visual-spatial abilities) that were below the Average
range and significantly lower than their WISC-III Vocabulary scores. These results are
more consistent with the findings of Shafrir and Siegel (1994) who compared pre- and
pst-secondary students with math disabilities and found that Group A Block Design
deficits were o d y present at the pst-secondary level. It should also be noted that unlike
in Rourke's studies, the current sample was not drawn from a clinical population.
Therefore, it appears that it is possible for grade 7 and 8 students to have a deficit o d y in
the area of mathematics and not display the visual-spatial difficulties that are thought to
be related to this academic profile.
Cornparison of the two MD subtypes on the dependent measures indicated with
the exception of the findings on the verbal STM task mentioned earlier, the A and ARS
Groups were not significantly different from each other on the other measures of memory
or on the processing speed or math concepts tasks. Thus the hvo subtypes were much
more alike than different on al1 of the cognitive and math measures assessed. Although
the Group A students with visual-spatial deficits found in other studies may be unique,
the MD Group in this study appears to be cornposed of faidy homogeneous subtypes.
This homogeneity suggests that similar remedial strategies could be used for both
subtypes.
If any trend emerged, it was that the Group ARS students were a weaker version
of the Group A students. Shafrir and Siegel (1994) also noted that their Group ARS
students seem to have more severe diiculties than their Group A students. Thus the
Group ARS and Group A students seem to be at different points on the sarne continuum.
This pattern is consistent with the normal variation view proposed by Willows (1991) and
Levine (198'7). Using this model, the MD Group would be seen as composed of subtypes
that are distinguishable by the intensity of theV difficulties rather than nature of their
difficuf ties.

A Potential Frameework for Math Compntation


The inclusion of a broad range of test measures thought to be related to math
computation has provided the opportunity to develop a more global profile of students
with math disabilities. When the current findings are combined with the results from
previous studies, a framework for math computation c m begin to be developed. The
schemata displayed in Figure 11 will be used to guide this discussion. Due to the fact
that there does not appear to be a unified theory of computation, and reading and math
appear to have many sirnilarities (Kulak, 1993)the theoretical frameworks available in
reading will also be used as a guideline.
The findings from this study indicated that students with math disabilities have
trouble with two aspects of computation, number fact retrieval and procedures. These
two aspects of computation are similar to automatic word recognition and the application
of phonetic rules in reading. Deficits in either of these areas of reading lead to more
reading errors, the expenditure of more energy to decode words and a reduced capacity to
focus on comprehension (Gaddes, 1985). The more often reading errors are made, the
weaker the association between the print and the word becomes and the Iess confident the
student is in hidher ability to read the word. This same pattern can be seen in math
computation.
Although the MD students appear to use similar back-up strategies as NC
students, they do not seem to make the shift to automatic number fact retrievai and when
they do, they tend to make more enors using this strategy. As a result, when the MD
students are completing computations, they make more calcuiation errors and they must
expend more energy using back-up number fact strategies. Therefore, when they are
coafronted with more complex questions (e-g.,those involving more steps, larger
numbers or that are embedded in word problems) the MD students have fewer rernaining
resources to deploy. The relationship between errors, confidence and automatic recall of
number facts are also evident in math and are addressed in Siegler's (1986)strategy
choice model.
In reading, a p o r visual memory is often the cause of the automatic retrieval
difficulties and poor phonological awareness hinders the development of phonetic skills.
There is also a relationship between these two areas, because the more phonetic patterns
that are recognized, the easier it is to recall words a u t o d c a l l y (Bruck, 1992).
Sequential memory is often implicated in both difficuities. A parailel pattern seems to
exist in math computation using the factors in the first row of the schemata in Figure 11.
The findings from the current study suggest that students in the MD Group had lower
mean scores than the NC Group in processing speed, basic math concepts and sequentiai
memory. Along with number fact automaticity, these factors were significant predictors
of computation ability. It appears that poor processing speed may contribute to the low
level of automatic retrieval in the MD Group and a poor grasp of basic concepts rnay be
related to procedurai errors. Basic math concepts also seem to influence the development
of back-up strategies for number fact retrieval. Although past studies have also
implicated working memory in number fact automaticity, no working memory
differences between the NC and MD Groups were found in this study. F i n d y , sequential
rnemory could be related to both number facts (e.g., counting strategies) and completing
procedural steps in the correct order. The potential cause and effect relationships
proposed above, along with the computation variables that were not addressed in this
study (those with a ? in Figure 1I), need to be investigated in subsequent studies, as the
current findings did not form associations between the factors in the first and second row,
but only between these factors and computation in general (third row) (represented by the
arrows in Figure 11).
Just as students with reading disabilities require both compensatory and remedial
strategies for their area(s) of weakness in order to be successful in reading, it seems that
students with math disabilities require compensatory and remedial strategies for their
area(s) of weakness in order to be successful in math. However, one of the differences
between these two academic areas, is that in math. each new ski11 buiids on the previous
one, whereas in reading, once decoding is mastered, there are no new decoding tasks. It
is this building block nature of math that leaves students with math disabilities
pdcuIarly Milnerable and tikely l a d s to the increased prevalence of math difficulties in
the higher grades.

Figure 11
Areas of Potential D=culty Within the Math Compatation Schemata

?
Number Facts Symbol Recognition
*

Computation
>

Contributions
This is the first study to investigate the memory, pmessing speed, number fact
skills and basic math concepts of one sarnple of students. As a result of this broad
exploration of skilis related to computation, the math computation difficulties of grade 7
and 8 students have been better delineated. Questions involving complex multiplication
and division and more particularly fractions, were the most difficuit for students in this
study to complete successfully. It was also determined that students with math
disabilities in computation tend to make the same types of errors as NC students, but they
make a larger number of calculacion and procedure errors. With knowledge about what
aspects of math computation seem to be most challenging and where the procedural
breakdowns occur, the cumculum could be designed to circurnvent these problems (e.g.,
more focus on the role of O as a place holder). Teacher's could also take a proactive roie
and try to address these pmblem areas before the arise.
The results from this study indicated that the MD Group has a poorer
understanding of basic math concepts than the NC Group. Although earlier studies have
suggested there may be some delays in this area in younger children, this is believed to be
the first study to look at basic math concepts in this age group. It appears that the delays
continue into the intermediate grades. These concepts are the foundation on which higher
level skills are built, thus it was not surprising to fmd that they contributed si-cantly
to the variability in math computation skills. Many of the procedural errors made by the
MD Group seemed to reflet weaknesses in these basic concepts. It dso appears that
without an understanding of the procedures, it is more difficult to generalize strategies
from one question to a similar question. Perhaps like poor readers who do not "see" the
underlying phonetic code, poor math students do not "see" the underlying number
patterns and d e s . If this is indeed the case, it seems intuitive that specZc teaching of
these basic math concepts would be beneficial.
With the inclusion of nurnber facts in al1 four operations, some overall patterns
became evident that had not been noted in studies of one operation at a tirne. The pattern
of automaticy was similar for both the NC and MD Groups and across al1 four operations,
suggesting that there may be one integrated retrieval network for dl four operations,
which is an extension of Campbell's proposal. For both groups and ail operations, the
ties were generally easier to answer than the non-ties. Some evidence to support the
hypothesis that this was due to the repeated operands was observed. The strategies the
students used to solve the number fact questions were also similar across groups and
operations. However, the NC Group retrieved signïficantly more number facts
automatically than the MD Group. Siegler's strategy choice mode1 (1986) was originally
designed for addition, but it seemed to be just as applicable to the other operations. The
rebleval strategy data also infonned Fuson et al.'s (199'7) mode1 about the development
of number fact strategies. Although the fmdings are not clear cut, it appears that number
fact automaticity may indeed be an area of defIcit for MD students. If this is the case, it
would seem prudent to provide these students with compensatory strategies, such as the
use of a calculator, so that they can focus their attention and energy on understanding the
question and applying the correct procedure.
Wiîh the exception of Bull and Johnson (1997), it is believed that no other math
studies include both rnemory and processing speed measures when investigating math
disabilities. Three interesting pieces of information came out of these analyses of
cognitive variables. First, no working memory differences were noted between the NC
and MD Groups, which was unexpected. Second,the two groups differed on measures of
processing speed. Therefore, perhaps processing speed as opposed to working memory is
one of the underlying difficulties in math computation, particularly as it relates to the
acquisition of number fact automaticity. Third, the content of the processing speed tests
seemed to be important, because when the tasks did not include nurnbers the more
conservative alpha level was not reached. Given that the inclusion of numbers bas also
been found to be important in past memory findings, it seems that numbers themselves
are a factor in math disabilities.
The similarity between the Group A and Group ARS students was one of the
surpnsing findings of the subtyping comparisons. The fact that most of the Group A
students did not display visual-spatial deficits was also unexpected, although this study
did not draw from a ciinical sample. Thus it appears that the some MD subtypes may be
more homogeneous than expected, with only the degree of their difficulties distinguishing
them from each other. Therefore, in this case it appears that similar remedid strategies
could be used for both subtypes of the MD Group. However, the subtype sarnple size
was small, so that there may be some hekrogeneity in the MD population at large that
was not captured in this study.
This is also one of the few math studies to include a regression analysis. It
appûars that number fact automaticity, basic math concepts and processing speed are
particularly important for the development of computation skills, but given the s d
sample size and the exploratory nature of this analysis, replication of the findings is
needed before firm conclusions can be made.
Overall, the students with math c o m p u t . o n disabilities in this study achieved
signif~cantlylower scores in basic math concepts, number fact automaticity, processing
speed and sequentid memory compared to the NC Group and made more procedural and
calculation errors. Combining these findings with those from other studies, it seerns that
the students with math disabilities are constantly trying to catch up with their peers in the
developrnent of computation skills and due to the building block nature of mathematics,
poor consolidation of one stage limits the success at the next stage.

Limitations and Futme Research


This study explored a number of different areas related to math computation in
grade 7 and 8 students. The Limited demographic information about the students is one of
the drawbacks to this study. No data was available about ethnicity, or SES (e-g., parents
education or income level). It should also be noted that unlike some of the studies which
draw from clinical samples of fomaily identified children, the students in this study were
identified as MD or NC based solely on their scores on the WFUT3 subtests. This may
limit the ability to generalize these fmdings to dl MD populations.
Al1 of the studies assessing computation abilities, including this one, identify
errors by analyzing the protocols after the fact. It would be interesting to interview
students as they complete computations to determine more clearly what procedures they
are using, where the breakdown occurs and what underlying beliefs it is based upon. In
this way remediation c m focus on the specific problem areas. Proactive steps could also
be taken by including expianations about the trouble spots when procedures are initially
being taught.
Many of the procedural errors the students made seemed to be related to place
value concepts, therefore a study exploring the relationship between these two variables
and the benefits of improved understanding of procedures is warranted. More research
like that being carried out by Hiebert, Carpenter and Weame, comparing curriculums that
include a focus on understanding versus those which focus on the mechanical aspects of
math also need to be carried out. A study by Markovitz and Sowder (1994) found that
estimation skills of a s m d l group of children could be improved through teaching.
Therefore it would be interesting to see if teaching estimation would help studeots with
MD and if the improved estimation skills would in turn make long division easier.
The majority of students in the present study used a mixture of strategies to
answer nurnber fact questions. The use of an automatic retrieval strategy was
significady Lower in the MD Group than the NC Group, even for those operations that
were intmduced earlier in the curriculum. A longitudinal study of changes in the strategy
mixes of students with math disabilities compared to normal controls would permit
cornparison of the developmental paths of these two groups and further enlighten the
debate between developrnentai delay and deficit The effects of remediation on these
strategy mixes would also provide evidence for the delay - deficit debate. In the current
study, the processing speed of the MD Group was si&icantly lower than that of the NC
Group. It would be interesting to determine if the use of calculators would bypass the
processing speed weakness, thereby permitting stronger operand-answer associations to
be fonned, resulting in improved number fact automaticity. The use of calculators could
aiso reduce the number of association errors that are made, producing less interference
and a more peaked association. Further research on the possibility of a single network
retrieval mode1 for nurnber facts is also needed, perhaps using response latencies in ali
four operations, to take the current fuidings one step further.
Continued investigation into MD subtypes is warranted, but perhaps from a
remedid point of view. One of the main reasons given for subtyping research is that
different rernediation may be needed depending on a student's patterns of strengths and
weaknesses. Thus the effects of remediation on skill development with different
subyptes may shed more light on whether the subtypes need different teaching methods
or not.
Replication of the results h m the current study related to processing speed and
working rnemory are needed, as the latter has been included in very few studies and the
results of the former were somewhat unusual. Replication is also needed of the findings
from the regression analyses. The current results hdicated that basic math concepts,
number fact automaticity and processing speed predicted approximately three quarten of
the variance in computation scores. However, which aspects of computation (e.p.,
number facts, procedure, symbol recognition) these variables are associated with was not
determined. Thus, the Iinks between the first and second row of the math computation
mode1 have yet to be established. With one quarter of the variance still unexplained there
is aiso room for other factors. The next step would be to include some of the
environmentai and emotional factors that may aiso be related to math computation (e-g.,
homework completion, attitude towards math,leamed helplessness, past learning
opportunities) in order to see if more of the variance can be explained. It is also
necessary to venture into other aspects of math such as geometry and word probtems, to
detemine how the profiles of strengths and weaknesses found in this study affect skill
acquisition in other areas of mathematics.
REFERENCES

Ackerman, P. T., & Dykman, R. A. (1995). Reading-disabled students with and


without comorbid arithmetic disability . Develo~mentalN e u r o ~ s v c h o ~ o 1~1.35
v , 1-371.

Ackerrnan, P. T., Dykman, R. A-, & Peters, J. E. ( 1976). Hierarcbid factor


patterns on the WSC as related to areas of learning deficit. Perce~tualand Motor Skills,
42,583-615-

Ackerman, P. T., Dykrnan, R. A., & Peters, J. E. (1977). Leaming disabled boys
as adolescents: Cognitive factors and achievement. Journal of the American Academv of
Child PsvcIùatrv, 16,296-3 13.

Ackerrnan, P.T., Anhalt, J. M., Dykman, R. A. (1%). Arithmetic


autornatization Mure in chiIdren with attention and reading disorders: Associations and
sequela. Journal of le am in^ Disabiiities. 19,222-232.

Andeason, K. E., Richards, H.C., & Hallahan, D.P. (1980). Ragetian task
performance of learning disabled children. Journal of Learninn Disabilities. l 3 , 5 O 1-505.

Ashcraft, M. H. (1992). Cognitive anthmetic: A review of data and theory.


Çoenition. 44,75106.

Ashcraft, M. H., & Battaglia, J. ( (19'78). Comitive aritbmetic: Evidence for


retrieval and decision orocesses mental addition. journal of Experirnental P s y c h o l o o ~
Human Leamin? and ~ e r n o r y4,527-538.
.

Ashcraft, M. H.,& Fierman, B. A. (2982). Mental addition in third, fourth, and


sixth grades. Journal of Experirnental Child Psychology. 33,216234.

Ashcraft, M. H., Yamashita, T. S., & Aram, D. M. (1992). Mathematics


performance in left and right brain-lesioned children. Brain and Cosmîtion. 19,208252.

Atkioson, B. (1983). Anthmetic remediation and the learning disabled


adolescent: Fractions and interest tevel. Journal of L.eamine Disabibties. 16,403-406-

Baddeley, A. D. (1981). The concept of working mernory: A review of its


current state and probable future development. -tien. 10,17-23.

Baddeley, A. D. (19%). Workin mernorv. New York: Oxford University Press.

Baddeley, A. D., Thomson, N.,& Buchanan, M. (19'75). Word length and the
structure of short-term rnemory. Journal of Verbal Leamins
- and Verbal Behavior. 14,
575589.
Badian, N. A. (1983). Dyscalculia and nonverbal disorders of learning. In H.R.
Mykelbust (Ed.), Promess in learning disabilities (Vol. 5, pp- 235-264). New York:
Stratton.

Beatty, L. S., Madden, R., & Gardner, E. F. (1966). Stadord Diagnostic


Anthmetic Test. New york: Harcourt, Brace & World.

Bednarz, N.,& Janvier, B. (1982). The understanding of numeration in primary


school. Educational Studies in Mathernatics. l3,33-57.

Benbow, C . P., & Minor, L. L. (1990). Cognitive Profiles of Verbally and


Mathematically Precocious Students: Implications for identification of the gifted. Gifted
Child Ouarterlv, 3421-25.

Bennett, G. K., Seashore, H. G., & Wesman, A. G. (1974). Manual for the
Differential Aptitude Tests. F o m S and T. New Y o r k The Psychological Corporation.

Bisanz, J., & LeFevre, J. A. (1990). Strategic and nonstrategic processing in the
development of mathematical cognition. In D. F. Bjorklund (Ed.), Children's stratemes:
C o n t e m p o q views of copitive development (pp. 213-243). Hilisdale, NJ: Erlbaum.

Bley, N. S., & Thomton, C. A. (19û9). T e a c h i n ~Mathematics to students with


leamine disabilities. Second Edition. Austin, TX: Pro-ed.

Branch, W. B., Cohen, M. J., & Hynd, G. W. (1995). Academic achievement


and attention-deficitmyperactivitydisorder in children with left- or right-hemisphere
dysfunction. Journal of Learnin~Disabilities. 28,35-43,64.

Brown, J. S., & Burton, R R. (1978). Diagnostic models for procedural bugs in
basic mathematical skills. Cognitive Science. 2,155 192.

Bruck, M. (1992). The word recognition and spelling of dyslexic children.


R e a d i n ~Research Ouarterl~.23,s1-69.

Bryant, B. R., & Pedrotty Rivera, D. (1997). Educational assessrnent of


mathematics skills and abilities. Journal of Learnin~Disabilities. 30,57-68.

Bull, R., & Johnson, R. S. (1997)- Children's arithmeticai difficulties:


Contributions from processing speed, item identification, and short-term memory.
Journal of Experirnental Child Psvchologv. 65, 1-24.

Campbell, J. 1. D. (1997). On the relation between skilled performance of simple


division and multiplication. Joumd of Ex erimental Psvcholo~v:Leaming. Mernorv,
and Coonition, 23, 1140-1 159.
CampbelI, J. 1. D., & Graham, D. J. (1985). Mental multiplication skill:
Structure, process and acquisition. Canadian Journal of Psvchologv. 39,338366.

Carpenter, T. P., Franke, M. L., Jacobs, V. R., F e ~ e m aE.,


, & Empson, S. B.
(1997). A longitudinal study of invention and understandimg in children's multidigit
addition and subtraction. Journal for Research in Mathematics Education. 29,3-20.

Carpenter, T. P., & Moser, J. M. (1984). The acquisition of addition and


subtraction concepts in -des one through three. Journal for Research in Mathematics
Education. 15,179-202.

Case, R. (198.5). Intellectual develo~ment:Birth to adulthood, San Diego, CA:


Academic Press.

Case, R., Kurkland, D. M., & Goldberg, J. (1982). Operational efficiency of


short term memory span. Journal of Ex rimental Ps~cholorrv.33,3ûfAW.

Cawley, J. F., fitunaunce, A. M., Shaw, R.,Kahn, H.,& Bates, H. (1979). LD


youth and mathernatics: A review of characteristics. Learnine Disabili
294.

Cawley, I. F., & Miller, J. H. (1989). Cross-sectional cornparisons of the


mathematical performance of children with learning disabilities: Are we on the right track
toward comprehensive programming? Journal of le am in^ Disabilitia. 22.250-259.

Cawley, J. F., Parmar, R. S., Yan, W. F., & Miller, J. H. (1996). Arithmetic
computation abilities of students with learning disabilities: Implications for instruction.
lem in^ Disabilities Research & Practice. 11,230-237.

Cipolotti, L,& de Lacy Costeilo, A. (1995). Selective impairment for simple


division. Cortex. 31,433-449.

Cohen, M. J. (1997). Children's Memory Scale. San Antonio, TX:The


Psychoïogical Corporation.

Comolly, A. J., Nachtman, W., & Pritchett, E. M. (1976).KevMath Diamostic


Arithmetic Test. Circle Pines, MN: American Guidance Service.

Connolly ( 1991). Canadian Edition of Key Math - Revised. Toronto, Ontario:


Psycm.

Cwney, J. B., Swanson, H. L., & Ladd, S. F. (1988).Acquisition of mental


multiplication skiU: Evidence for the transition between counting and retrieval shlitegies.
Cognition and Instruction. 5.323-345.
Cowan, N. (1992). Verbal memory span and the timing of spoken recall. Journal
of Memory and L a n ~ u a ~3e1,66&.684.
.

Cowan, N., Keller, T.A., Hulrne, C., Roodenrys, S., McDougaIl, S., & Rack, J.
(1994). Verbal memory span in children: Speech timing clues to the rnechanisrn
underlying age and word length effects. Journal of Memorv and Laaguage, 33.234250.

Cumming, J. J., & EIkins, J. (1994). Are any errors careless? Focus on Leamine
Problems in Mathematics. 16(4), 21-30.

Davis, R.B., & McKnight, C. (1980). The influence of semantic content on


algorithmic behavior. Journal of Mathematical Behavior. 3,3943'7.

DeBenencourt, L. U.,Putnam, R T., & Leinhardt, G. (1993). Learning disabled


students ' understanding of derived fact strategies in addition and subtraction. Focus on
Laming Problems in Mathematics. 15,2743-

De Corte, E., & Verschaffel, L. (1981). Children's solution process in


elementary &thmetic problems: analysis and improvement. Journal of Educational
PsvchoIoev. 73,765-779.

Derr, A. M. (1985).Conservation and mathematics achievement in the learning


disabled child. Journal of Leamin9 Disabilities. 18,333-336.

Dunn, L. M., & Dunn, L. M. (1981) Peabody Picture Vocabularv Test - Revised.
Circle Pines, MN: American Guidance Service, Inc.

Eiley, W. B., & Reid, N.A. (1%9). Promessive Achievement Test, teacher's
Manual. R e a d i n e c o m r > r e h e n s i o n . Vocabul . Wellington: New Zealand
Council for Educational Research.

Eüiott, C. D. (1990). Differential Abilities Scale. San Antonio, TX:The


Psychological Corporation

Ellis, N. C., & H e ~ e l l e yR.


, A. (1960). A bilingual word length effect:
Implications for intelIigence testing and the relative ease of mental calculation in Welsh
and English. British Journal of Psvcholoev. 71,169-191.

Engelhardt, J. M. (1977). Analysis of children's computational errors: A


qualitative approach. British Journal of Educational Psvchology. 47,149-154.

Enright, B. (1985). Mastery of math facts: Activities that help. Childhood


Education, 62. 148-150.

Ericsson, K. A., & Kintsch, W. (1995). Long terrn working memory.


Ps~cholo@cal Review. 102,211-245.
Ericsson, K. A., & Simon, H. A. (19%). Rotocol analysis: Verbal g o r t s as
-
data. Cambridge, MA: MIT Press.

Feogens, L., & McKimey, J. D. (1981). The pattern of exceptionality across


domains in leaming disabled children. Journal of Applied Developmental Ps~choloev.1.
3 13-328.

Reischner, J. E., Garnett, K., & Shepherd, M. J. (1982). Proficiency in


arithmetic basic fact computation of leaming disabled and nondisabled children. Focus
on Learning Problems in Mathernatics. 4,47-56.

Fleishner, J. E, & Manheimer, M. A. (1997). Math interventions for students


with learning disabilities: Myths and realities. School Psvcholow Review. 26,3W413.

Retcher, J. M. (1985). Memory for verbal and nonverbal stimuli in learning


disability subgroups: Analysis by selective reminding. Journal of Experimentai Child
Psycholoev. 40,244-î59.

Fuson, K. C. ( 1992). Research on learning and teaching addition and subtraction


of whole numbers. In G. Leinhart, R. futnam, & R. A., Hattnip (Eds.), AnaIysis of
arithrnetic for mathematics teaching (pp. 53-187). HiIlsciale, NJ: Erlbaum.

Fuson, K.C. (1990). Conceptual structures for multiunit numbers: Implications


for leaming and teaching multidigit addition, subtraction, and place value. Comition and
Instruction, 7,343403.

Fuson, K. C., & Briars. D. J. (1990). Using a base-ten blocks leaming/teaching


approach for first- and second-grade place value and multidigit addition and subtraction.
Journal for Research in Mathematics Education, 21,180-2%.

Fuson, K. C., Stigler, J. W., & Bartsch, K. (1986). Grade placement of addition
and subtraction topics in Japan, Mainland China, the Soviet Union, Taiwan,and the
United States. Journal for Research in Mathematics Education, 19,449-456.

Fuson, K. C., Weame, D., Hiebert, J. C., Murray, K.G., Human, P. G., Olivier, A.
1.. Carpenter, T.P., & Fennema, E. (1997). Childreo's conceptual structures for
multidigit numbers and methods of multidigit addition and subtraction. Journal for
Research in Mathematics Education. 28,130-162.

Gaddes, W. H. (1985). The neuropsychological bases of problems in wnting,


reading, and spelling. In W. H. Gaddes (Ed.1. Le-ne disabilities and brain function: A
neuroosvcholoPical a~proach(pp. 328-369). New York: Springer-Verlag.
Gmett, K., & Fleischner, J. E. (1983). Automatization and basic fact
performance of nomal and learning disabled children. Leamino Disabilities Ouarterlv,
6,223-230.

Gathercole, S. E., & Adams, A. (1994). Children's phonological working


memory: Contributions of long term kuowledge and rehearsd. Journal of Memorv and
Lan ~ u a ~33,672-688.
e.

Geary, D. C. (1990). A componential analysis of an early learning deficit in


rnathematics. Journal of Exoerimental Child Psvchology-49,363-383.

Geary , D. C. ( 1993). Mathematical disabilities: Cognitive, neuropsychological,


and genetic cornponents. Psycholoeicd Bulletin. 1 14,345-362.

Geary, D. C. (1994). Children's mathematical develovment Washington, DC:


American Psychological Association.

Geary, D. C. (19%). The problem-size effect in mental addition: Developmental


and cross-national trends. Mathematical Cognition. 2,63-93.

Geary, D.C., Bow-Thomas, C. C., & Yao, Y. (1992). Counting knowledge and
skill in cognitive addition: A cornparison of normal and mathematicaily disabled
children. Journal of Ex~erimenta.1ChildPsvcholo~v.54,372-391.

Geary, D. C., & Brown, S. C. (1991). Cognitive addition: Strategy choice and
speed-of-processing differences in gifted, normal and mathematically disabled children.
Developrnental Psvchologv. 27,3984U6.

Geary, D. C., Brown, S. C., & Samaranayake, V. A. (1991). Cognitive addition:


A short longitudinal study of strategy choice and speed-of-processing differences in
normal and mathematically disabled children. Develo~mentalPs chology. 2 7 m 787-
7!37.

Geary, D. C., & Burlingham-Dubree, M. ( 1989). Extemal validation of the


strategy choice mode1 for addition. Journal of Emerimental Child Psvcholo~v.47,175
192.

Geary, D.C., & Wideman, K. F. (1987). Individual differences in cognitive


arithmetic. Journal of Experirnental Psycho10 V: General. 116, 154-171.

Geary, D. C., & Widaman, K. P. (1992). Numerical cognition: O n the


convergence of componential and psychometric rnodels. Intelli~ence.16,47430.

Geary, D. C., Widaman, K. F., Little, T. D.,& Courrnier, P. (1987). Cognitive


addition: Cornparison of leaming disabled and academically normal elementary school
children. Comitive Development, 2,249-269.
Ginsburg, H.P. (19û7). The development of arithmetic thinking. In D. D.
Hammill (Ed),as ses sin^ the abilities and instructional needs of shidents (pp. 423-440).
Austin, TX:PRO-ED.

Ginsburg, H. P. (1989). Children's arithmetic: How thev l e m it and how vou


teach it (Znd ed.). Austin, TX: Pro-ED.

Ginsburg, H.P. (1997). Mathematics learning disabilities: A view from


developmental psychology. Journal of Leamine Disabilities. 30,20-33.

Goldman, S. R., Pellegrino, J. W., & Mertz, D. L. (1988). Extended practice of


basic addition facts: Strategy changes in leaming disabled students. Comition and
Instruction, 5,223-265. -

Goldman, S. R., Mertz, D. L,& Pellegrino, J. W. (1989). Individual diMerences


in extended practice functions and solution strategies for basic addition facts. Journal of
Educational Ps~chologv.81.4814%.

Graham, D.J., & Campbell, J. 1. D. (1992). Network interference and number-


fact retrieval: evidence from children's alphaplication. Canadian Journal of Psvcholoey,
46.65-9 1.

Greeno, J. G. (1991). Number sense as sitmted knowing in a conceptual domain.


Journal for Research in Mathematics, 22. 170-218.

Greenstein, J., & Strain, P.S. (197ï). The utility of the Key Math Diagnostic
Anthmetic Test for adolescent learning disabled students. Psvchotogv in the Schools. 14,
275-282.

Gross-Tsur, V., Shalev, R. S., Manor, O., & Amir, N. (1995). Developmenial
right-hemisphere syndrome: Clinical specburn of the nonverbal learning disability.
Journal of Learning Disabilities. 28, ûû-86.

Hamann, M. S., & Ashcraft, M. A. (1986). Textbook presentations of the basic


addition facts. Comition and Instruction. 3, 173- 192.

Hamadek, M. C. S., & Rourke, B. P. (1994). Principal identifying features of the


syndrome of nonverbal learning disabilities in children. Journal of Leamin? Disabilities,
27,144-154.

Hasher, L., & Zacks, R T. (1979). Automatic and effortful processes in memory.
Journal of Experimental Psvcholo~v:General, 108,356-388.

Henry, L. A., & Miller, S. (1991). Mernory span increases with age: A test of
two hypotheses. Journal of Expenmental Child Psvcholog. 51.459484.
Hiebert, J., & Carpenter, T. P. (1992). Learning and teaching with
understanding. In D. A. Grouws (Ed.). Handbook of research on mathematics teaching
and learning (pp. 6597). New York: Macmillan.

Hiebert,J., & Lefevre, P. (1986). Conceptual and procedural knowledge in


mathematics: An introductory anaiysis. In J. Hiebert (Ed.). Conceptual and ~rocedural
knowledae: The case of mathematics (pp. 1-27). Hillsdale, NJ: Lawrence Erlbaum
Associates, Inc.

Hiebert, J., & Weame, D. ( 1992). Links between teaching and leaming place
value with understanding in first grade. Journal for Research in Mathematics Education,
23,98-122.

Hiebeit, J., & Wearne, D. (1996). Instruction, understanding, and ski11 in


multidigit addition and subtraction. Çopnition and Instruction. 14-251-283.

Hitch, G. J., Halliday, M. S., & Littler, J. E. (1989). Item identification time and
rehearsal as predictors of memory span in children. Quarterly Joumal of Experirnental
Psvcholo~v.41A, 321-338.

Hitch, G. J., & McAuley, E (1991). Working mernory in children with specific
arithmetic leaming difficulties. British Journal of Psvcholo~v.82.375386.

Hittmair-Delazer, M., Semenza, C., & Denes, G. (1994). Concepts and facts in
calculation. Brain, 1 17,715-728.

Howell, R, Sidorenko, E., Junca, J. (2987). The effects of cornputer use on the
acquisition of multiplication facts by students with learning disabitities. Journal of
Leaminp Disabilities. 20,336-341. (1987).

Hulme, C., Maughan, S., & Brown, G. D. A. (1991). Memory for famüiar and
unfamiliar words: Evidence for long-term memory contributions to short-terni memory
span. Journal of Memon, and Lanmaee. 30.373-398.

Jastak, 3. F., & Jastak, S. R. (1965). The Wide Rawe Achievement Test.
Wilrnington, DE: Guidance Associates.

Jastak, S., & Wilkinson, G. S. (1984). The Wide Range Achievement Test -
Revised. Wilmington, D E Jastak Assotiates Inc.

Johnson, D. J., & Mykiebust, H. R. (1967). Learning: disabilities: Educational


principles and practices. New York: Grune & Stratton.
Jordan, N. C., & Oettinger Montani, O. (1997). Cognitive arithmetic and
problem solving: A cornparison of children with specifc and general mathematics
difficulties- Journal of Learninp Disabilities. 6,624434,684.

Kail, R. (1992). Developmentai functions for speeds of cognitive processes.


Journal of Ex~enmentalChild Psvchology. 45.339-364.

Kaye, D.B., Post, T. A., Hall, V. C., & Dineen, J. T. (1986). Ernergence of
information-retrievd strategies in numerical cognition: A developmental study.
Cognition and Instruction. 3,127- 150.

Kaye, D. B., deWinstardey, P., Chen, Q., & Bonnefil, V. (1989). Development
of efficient arithmetic computation. Journal of Educationai Psvcholo~v.81.467480.

Kosc, L. (1974). Developmentd dyscalculia. Journal of Learnin~Disabilities. 7,


165-177.

Kouba, V.L., Brown, C.A.,Carpenter,T. P.,Linquist, M. M., Siiver, E. A.,&


Swafford, J. 0. (1988). Results of the fourth NAEP assessrnent of mathematics:
Number, operations, and word problems. Anthmetic Teacher. 35(8), 1 4 19.

Kulak, A. G. (1993). Parallels between math and reading: Common issues and
approaches. Journal of Learning Disabilities. 26,666-673.

LeFevre, J. -A., Bisanz, J., Daley, K. E., Buffone, L,Greenham, S. L., &
Sadesky, G. S. (1996). Multiple routes to solution of single-digit multiplication
problmes. Journal of enmental Psvchoinov: General, 125,284-306.

Lemaire. P., & Siegler, R S. (1995). Four aspects of strategic change:


Contributions to children's leaming of multiplication. Journal of Experimental
Psvcholo~y:General, 124.83-97.

Levine, M. D. ( 1987). Developmental variation and ZearninP disorders.


Cambridge, MA: Educators Publishing Service, inc.

Light, J. G., & Defries, J. C. (1995). Comorbidity of readingand mathematics


disabilities: Genetic and environmental etiologies. Journal of Leaniing Disabilities. 28,
5106.

Logie, R. H., Gilhooly, K. J., & Wyun, V. (1994). Counting on working memory
in arïthmetic problem solving. Memorv and Copnition. 22.395410.

Markovits, Z.,& Sowder, J. (1994). Developing number sense: An intervention


study in grade 7. Journal for Research in Mathematics Education. 25,4-29.
McLeod, T. M., & Armstrong, S. W. (1982). Learning disabilities in
mathematics-ski11 deficits and remedial approaches at the intermediate and secondary
level. Leaniin~Disabilitv Ouarterlv, 5,305311.

McLeod, T. M., & Cmmp, W. D. ( 1978). The relationship of visuospatial skiils


and verbal ability to learning disabilities in mathematics. Journal of Leaming
Disabilities. 11,237-241.

Miller, J. H., & Milam, C. (1987). An examination of error patterns arnong the
l e d n g disabled. Learnin~Disabilities Research. 2,119- 122.

Miura, 1. T., & Okamoto, Y. (1989). Cornparisons of U.S. and Iapanese fmt
=ders' cognitive representation of number and understanding of place value. Journal of
Educationd Psvcholow, 81,109- 114,

Naglian, S. A., & Das, J. P. (2997). Cornitive Assessment S~stem.Itasca, IL:


Riverside Publishing.

Nolan, D. R.,Harnmeke, T. A., & Barkiey, R. A. (1983). A cornparison of the


patterns of the neuropsychologicai performance in two groups of leaming disabled
chitdren. Journal of CLinical Child Psvcholo~v.12,22-27.

Norman, C. A., & S igmond, N (1980). Characteristics of children labeled and


served as leaming disabled in school systems affiliated with the Child Service
Demonstration Centers. Journal of Learning Disabilities, 13,542-547.

Ohlson, S., & Rees, E. (1991). The function of conceptual understanding in the
leaming of arithmetic procedures. Cognition and Instruction. 8,103- 179.

Ozols, E. J., & Rourke, B. P. (1988). Characteristics of young learning-disabled


children classifïed according to patterns of academic achievement: Auditory-perceptual
and visual-perceptual disabilities. Journal of Clinical Child P s ~ c h o l o g 17.44-52.
~.

Rashen, P., & Ahonen, T. (1995). Arithmetic disabilities with and without
reading difficulties: A cornparison of xithmetic errors. Developmental
Neurops~cholo~v. 11,275-295.

Raven, J. C., Court, J. H., & Raven, J. (1977). Raven's Coloured Propessive
Matrices. London, England: J. C. Raven Ltd.

Reid, N. A., & Hughes, D. C. (1974). Promessive achievement tests. teacher's


manual. Mathematics. Wellington: New Zealand Council for Educational Research.

Reitan, R. M. (1966). A research program on the psychological effects of brain


lesions in human beings. In N.R. B l i s (Ed.),International review of research in mental
retardation (pp. 153-218). New York: Academic Press.
Resnick, L. B. (1983). A developmental theory of number understanding. In H.
P. Ginsburg (Ed.), The develo~mentof mathematical thinking (pp. 109-151). San Diego,
CA: Academic Press.

Reys, R. E-, Rybolt, J. F., Bestgen, B. J., & Wyatt, J. W. (1982). Processes used
by good c&nputationzd estimators. ~oÜrnalfor ~ e s e a k hin Mathematics Education. 13,

Rickard, T. C., Healy,A. F., & Bourne, L. E. Jr. (1994). On the representation of
arithmetic facts: Operand order, symbd, and operation transfer effects. Journal of
Experimentaf Psycho10 Y: General. 117,258.275-

Roberts, G. H. (1968). The failure strateses of third grade arithmetic pupils.


Arithmetic Teacher. 15,442-446.

Roodenrys, S., Hulme, C., & Brown, G. (1993). The developrnent of short term
memory span: Separable eifects of speech rate and long term memory. Journal of
Exoerirnental Child Psvchologv. %,43 1-442.

Rourke, B. P. (1993) Arithmetic Disabilities, specific and otherwise: A


neuropsychological perspective. Journal of Leamine Disabilities. 26.214-226.

Rourke, B. P. & Conway, J. A. (1997). Disabiities of arithmetic and


mathematical reasoning: Perspectives from neurology and neuropsychology. Journal of
Leamin9 Disabilities, 3 0 , 3 4 4 6 .

Rourke, B. P., & Finlayson, M. A. J. (1978). Neuropsychological ~ i ~ c a nofc e


variations in patterns of academic performance: Verbal and visual-spatial abilities.
Journal of Abnormal Child Psycholo~v.6,121-133.

Rourke, B. P., & Strang, J. D. (1978). Neuropsychological significance of


variations in patterns of academic performance: Motor, psychomotor, and tactile-
perceptual abilities. Journal of Pediatric Psycho10 w.3,6246.

Rubenstein, R. N. (1985). Computational estimation and related mathematical


skills. Journal for Research in Mathematics Education. 16, 1 6 119.

Rwell, R. L., & Ginsburg, H. P. (1984). Cognitive analysis of children's


mathematics difficulties. Coonition and Instruction. 1,217-244.

Saito, S. (1992). Ouantitative and qualitative analvsis of arithmetic errors of


arithmetic disabled and normallv achievino children. Unpublished master's thesis,
University of Toronto,Toronto, Ontario, Canada.
Sattler, J. M. (1992)- Assessrnent of children: Revised and u~datedthird edition.
San Diego: Jerome M. Sattler, Publisher, Inc.

Saxe, G. B., & Shaheen, S. (1981). Piagetian theory and the atypical case: An
analysis of the development of Gerstmann's syndrome. Journal of Learnine Disabilities,
14,131-135-

Sutaria, S. D. (1985).Specific leamine disabilities: Nature and needs.


Springfield, IL:Charles C.Thomas.

Shafrir, U-,& Siegel, L. S. (1994). Subtypes of learning disabilities in


adolescents and adults. Journal of Learning Disabilities, 27,123-134.

Share, D. LI,Moffit,T. El,& Silva, P. A. (1988). Factors associated with


arithmetic-and-reading disability and specific arithmetic disability. Journal of Learning
Disabilities. 21,3 13-320.

Siegel, L. S., & Heaven, R. K. (1986) Categonzation of learning disabilities. In


S. J. Ceci (Ed.),Handbook of comitive. social. and neurolo~calas~ectsof learning
disabilities, Vol. 1. Hillsdale, Nk Etlbaum.

Siegel, L. S., & Linder, B. A. (1984). Short-term memory processes in children


with reading and arithmetic tearning disabilities. Developmental Psvcholo~v.20,200-
207.

Siegel, L. S., & Ryan, E. B. (1988). Development of grammatical-sensitivity,


phonological, and short-terni memory skills in nomally achieving and learning disabled
chüdren. Developmental Psvchology. 24,2837.

Siegel, L. S., & Ryan, E.B. (1989). The development of working memory in
normally achieving and subtypes of learning disabled children. Child Development. 60,
973-980.

Siegler, R. S. (1986). Unities across domains of children's strategy choices. In


M. Perlrnutter (Ed.), Pemectives for intellecrual develo~ment:Minnesota svrnposia on
child psycholog (Vol. 19,pp. 1-48).Hiilsciale, NJ: Erlbaum.

Siegler, R. S. (1987). The perils of averaging data over strategies: An example


from children's addition. Journal of Ex~erimentdPsycholo Y: General. 116,250-264.

Siegler, R. S. (1988). Individual differences in strategy choices: Good students,


not-so-good students, and perfectionists. Child Develomnent. 59,833-851.

Siegler, R. S. (1989). Hazards of mental chronometry: An example from


children's subtraction. Journal of Educational Ps~cholor~v.
81,497-506.
Siegler, R. S., & Robinson, M. (1982). The development of numerical
understanding. In H. Reese & L. P. Lipsitt (Eds.), Advances in child develooment and
behavior (Vol. 16, pp. 241-3 12). San Diego, CA: Academic Press.

Siegler, R. S., & Shrager, J. (1984). Strategy choice in addition and subtraction:
How do children know what to do? In C. Sopbian (Ed.),Orioins of cosmitive skills (pp.
229-293). Killsdaie, NJ: Erlbaum.

Spiers, P. A. (1987). Acalculia revisited: Current issues. In G. Deloche & X


Seron (Eds.), Mathematical disabilities: A cognitive neuropsvchological perspective
(pp. 1-26). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Steinberg, R M. (1985). Instruction on denved facts strategies in addition and


subtraction. Journal for Research in Mathematics Education, 16,337-355.

Strang, J. D., & Rourke. B. P. ( 1 8 ) . Concept-formationhon-verbal reasoning


abilities of children who exhibit specifc academic problems with anthmetic. Journal of
5 .
12,33-39.

Strauss, A. A., & Lehtinen, L. E. (1947). Psychopathology and education of the


brain-injured child. NY: Grune and Stratton.

Svenson, O., & Broquist, S. (1975). Strategies for solving simple addition
problems: A cornparison of nomal and subnormal children. Scandinavian Journal of
Ps~chology.16,143-151.

Swanson, H. L. (1993). Working memory in learning disability subgroups.


Journal of Leaming Disabilities. 20,3-7.

Swanson, H. L. (1994). The role of working memory and dynamic assessrnent in


the classification of children with learning disabilities. Leamin? Disabilities Researcb
and Practice, 9,190-202.

Swanson, H.L., Cochran. K.F.. & Ewers, C.A. (1990).Can learning disabilities
be determined from working memory performance? Journal of Leamino Disabilities. 23,
59-67.

Tabachnick, B. G., & Fideil, L. S. (1983). us in^ multivariate statistics. New


York: Harper & Row .

Temple, C. M. (1991). Procedural dyscalculia and number fact dyscalculia:


Double dissociation in developmental dyscalcutia. Coenitive Neuropsvcholo~v.8,155
176.
Thompson, A. G. (1992). Teacher's beliefs and conceptions: A synthesis of the
research. In D. A. Grouws (Ed-), Handbook of research in mathematics teachiw- and
IeamÎng (pp. 127- 146). New York Macmillan.

Threadgill-Sowder, J. (1984). Computational estimation procedures of schwl


children. Journal of Educationd Research. 77,332-336.

Tracy, D. M. (1994). Using mathematical ianguage to enhance mathematicai


concephialization. Childhood Education, 70,22 1-224.

Turner, M. L., & Engle, R. W. (1989). 1s working rnemory capacity task


dependent? .Tomal of Memow and L a n ~ u a ~28.127-
e. 154.

Webster, R. E. (1979). Visual and short-terni memory capacity deficits in


mathematics disabled students. Journal of Educational Research, 72,277-283.

Wechsler, D. (1991). Wechsler Intellkence Scale for Children - Third Eidition.


San Antonio, TX: The Psychological Corporation.

Weinstein, M. L. (1980). A neurological approach to math disability. New York


Universi- Education Ouarterly. 11.22-28.

Wikinson, G. S. (1993). Wide Rawe Achievement Test - Revision 3.


Wilrnington, DE: Wide Range Inc.

Willows, D.M. (1991). A "normal variation" view of wntten language


difficulties and disabilities: Implications for whole language prograrns. Exceptionality
Education Canada, 1(3), 1-31.

Woodcock, R. W. (1997). Woodcock Diamiostic Readinp Batterv. Itasca, IL:


Riverside Pubiishing.

Woodcock, R. W., & Johnson, M. J. (1977). Woodcock-Johnson


Psvchoeducational Assessrnent Batterv. Allen, TX: DLM.
APPENDIX A

CONSENT FORM

Letter of Permission
Dear Parent or Guardian,

1 am in a Ph-D. probofam at the University of Toronto. As part of that program I


am carrying out research on how children in grades 7 and 8 l e m math. The principal at
(name of schooi), (name of principal). the X Board of Education and the University of
Toronto have given permission for this study to take place at (name of schooi). This
letter is to tell you about the study and invite your child to participate in it.
The study will include grade 7 and 8 students who have good math computation
skills as weil as those who have dficulties in this area. Along with assessing the
students' skill levels in a variety of areas, four remedial sessions will be offered to some
of the students with math weaknesses. It is hoped that the results of tùis study will
contribute to a better understanding of math computation skills and how they can be
remediated. No information about an individual student will be shared with the teacher,
you or your child. Al1 students will be given a list of math activities and resources that
could be used at home. The study is divided into two stages.
Stage 1
In the f m t stage the students will be given a 15-minute math and spelling test in
class. Some of those students will be selected to take part in stage 2.

Stage 2 (you will be contacted if your child is involved in stage 2)


In stage 2 the students wiU complete math tests that assess a variety of different
math skills. Along with the diiferent math tests, each child's learning potentiai, mernory,
visual-spatial and motor skills will be lwked at, as these skills seem to be related to math
development Reading and spelling tasks are also included to see if children who only
have trouble in math learn differently than children who have trouble in reading, spelling
and math. Attempts will be made to do this testing during lunch or after school. It
should take approximately one hour to complete. Testing wiU always be stopped if a
child appears unhappy or overly tired.
Some of the students who had trouble with the first stage math test will also be
given four 45 minute session of remediation to review and try to improve their addition,
subtraction, multiplication and division skills.
The study will take place from February to May during this school year. The
results from al1 of the participating students wilI be analyzed and wntten up in my
doctoral thesis. No information that could identify an individual child or school will be
disclosed in the write up. Participation in this study is purely voluntary. You or your
child may choose to withdraw from the study at any time with no negative effects. Please
discuss the study with your child and choose whether or not you want to participate by
marking yes or no on the Letter of Agreement, simning your names and returning the
letter to the school. If you would like a surnrnary of the overall study resuits please dso
fil1 out that section on the Letter of Agreement. If you have any additional questions
about the study, please contact me at 396-7929.Thank you.

Cynthia McCaii, MA. Dale Willows, Ph.D., C. Psych.


Psychoeducational Consultant Doctoral Supervisor
(phone number) (phone number)

Letter of Agreement

Yes,1 agree to have rny child participate in this math


study. (name)

No, I do not agree to have my child participate in


this math study. (name)

1 understand that if this study is published, no information will be used that would
identify an individual child or hisher school.

Parent Name : Student Name :


Signature: Signature:
Date:
Phone Number:
INVESTIGATOR CREATED ESTIMATION AND PLACE VALUE TESTS

Estimation Test
Five questions were presented to the students. They had to esîimate the answer to each
questions without using a paper and pencil.
Two questions (1 1 and 15) from the KeyMath- R Estimation subtest
2 4 x 19= (acceptable range was 400 - 500)
497 t 27 = (acceptable range was 15 - 25)
3268 i 45 = (acceptable range was 643 - 90)

Place Valae Test


The students were shown one question at a tirne and asked to provide their answer oraily.

What digit is in the hundreds place? 82,369


What digit is in the thousamis place? 4,785,623
Round to the nearest hundredth? 2.6237
576 = 4 hundreds, ? tens, 6 ones
APPENDIX A

INVESTIGATOR CIUEATED STAGE 1 NUMBER FACT TESTS

Stage 1 Nnmber Fact Test - Addition

M e n 1 Say go, start at the top of the List, and going in order, without skpping any, keep
working untilI Say stop. Work as quickiy as you can without making mistakes.
APPENDIX A

-
Stage 1Nomber Fact Test Snbtraction

When 1 Say go, start at the top of the List, and going in order, without skipping any, keep
working until 1 say stop. Work as quickly as you can without making mistakes.
APPENDIX A

-
Stage 1Number Fact Test Multiplication

When 1 Say go, start at the top of the List, and going in order, without skipping any. keep
working until 1 Say stop. Work as quickly as you can without maklng mistakes.
-
Stage 1 Niimber Fact Test Division

When I Say go, start at the top of the ILS& and going in order, without skipping any, keep
working until I Say stop. Work as quickly as you c m without making mistakes.
LNVESTIGATOR CREATED STAGE 2 NUMISER FACT TESTS

Stage 2 Number Fact Tests - Addition and Subtraction

ADD SUBTRACT

7+8= 14-7=

6+6= 15-9=

9+7= 16-8=

8+6= 13-7=

9+9 = 18-9=

7+6= 15-7=

8+9= 16-9=

7+7= 14-6=

6+9= 17-8=

8+8= 12-6 =
APPENDIX A

Stage 2 Nnmber Fact Tests - Muitipfication and Division


LNVESTIGATOR CREATED RAPID NUMBER NAMLNG TEST

The students were presented with a card with nine numbers on it. They were
asked to read the nurnbers in each row as quickly as possible without making any
mistakes and to continue until all of the numbers had been rad. This was done with both
single-digit and two-digit numbers, as the operands and answers of number fact questions
include both types of numbers.

Single-Digit Numbers

Two-Digit Numbers

32
Level of Aotomaticity and Rankings for all Four Operations
Tables B 1 to 84 display the percent of automatic retrieval and rankings for each
of the number facts presented in Stage 2 of this study

Table B1

Percent of Each Addition Question Answered Antomatidy


Addition MD NC
Percentage 1 Rank Percentage 1 Rank

Table B2
Percent of Each Subtraction Question Answered Automatidy
Snbtraction 1 MD 1 NC 1
Percentage Rank ~ e r c e n t a ~ e Rank
12-6s 90.0% 1 933% 4
Table B3

Percent of Each Mdtiplication Question Answeed Automaticaily


Multiplication MD NC
Percentage Rank Percentap 1 Rank
6x6=36 90.0% 1 100.0% 1
6x742 333% 6 96.6% 5

Table FM

Percent of Each Division Question Answered AutomaticalIy

Percentage Rank Percentage Rank


36&==== 83.3% 1 100.0% 1
4267 433% 5 100.0% 1
&+6S 30.0% 8 833% 10
4
4
7
-
9
4
7
I

Add Subtract Multiply Add Subtract Multiply


Add
Subtract .72
Multiply -85 -69
Divide -73 -35 .79

Two-tailed test r = .87,p < .ml


NUMBER FACT ERRORS

These tables display the total number of errors made by each group for each
operation. The mean number of total number fact errors cornmitted the NC Group was
1 .W(SD = 1.63). The mean number of total number fact errors cornmitted be MD Group
was 4.17 (SD= 3 S5). There was wide variability within the MD Group. Only two
students made no enors, twenty studeots made i - 5 errors, six students made 6 - 10
errors and two made more than 10 erron. Therefore, approximately two thirds of the
students were quite effective at selecting strategies that would produce the correct
answer.

Table B5

Total Nuinber of Errors for Each Opemtion by Groop


Group Addition Subtraction Multiplication Division Tota1

NC 6 27 11 14 58

MD 15 41 45 24 12.5

The proportion of erron was also calcdated. The MD Group did not attempt al1
of the questions in multiplication and division, either because they did not know how to
answer a question or their strategy was so time consurning that they explained how they
would get the answer, but did not achially carry out the procedure.

Table 336

Proportion of Errors for Each Operation by Gmnp (in percent)


Group Addition Subtraction MultipIication Division Total

NC 2 9 4 5 5

MD 5 14 19 Il 12
APPENDIX B

NUMBER FACT ERRORS BY STRATEGY

The following tables display the nurnber of erron each group made using each
type of strategy. The fables read as follows: the NC Group made errors on 5 of their 277
automatic retrieval responses and on 1 of their 6 counting responses. Inspection of the
data sugpsts that the MD Group &es more emrs on back up strategies than on
automatic retrîevaI. The proportion of errcrs seems to be similar for both the NC and
MD Groups in addition (1.8%and 0.6%, respectively), subtraction (7.5% and 9.6%,
respectively) and division (4.6% and 6.796, respectively). In contrast, the NC Group
seems to make relatively smailer number of errors than the MD Group in multiplication
(3.9% and 11.8951, respectively).

Table B7

Addition Errors by Strategy


Group Auto Slow Auto Reference Counting

NC 9277 O12 0/15 1/6

MD 1/179 0/2 3/42 5/77

Table BS

Snbtraction Errors by Strate=


Group Auto Slow Auto 1 Reference Counting

NC 18/240 2/11 5/25 2/24

MD 13/ 135 O/1 2/30 î7/ 134


I
Table B9

Multiplication Errors by Sîrategy


Auto Slow Auto 1 Reference 1 Counting

Table 810

Division Emrs by Strategy


Group Auto Slow Auto Reference Counting

NC 13/285 1/6 0/9 O/O

MD 9/134 O/ 12 13/63 3/32


SUBTRACTION -ER FACT ERRORS

The number of subtraction counting up and counting down errors are presented in
terms of the total number of times each of these strategies was used to determine the
answer to a number fact question. I t appears that more errors are made using the
counting down strategy.

Table BI1

Subtraction Counting Up and Counting Down E r r o ~


Group Count up Count Down
The type of multiplication errors made by students on the Stage 2 number facts
was catculated for both groups. Table related errors are those in which the answer is
correct for one of the operands (e.g., 6 x 7 = 48 is correct for the 6 times table). A close
error had to be within ten percent of the correct answer. In two of the cases in the MD
Group, the close errors were in a multiplication table. but not one retated to the operands
(e.g., 7 x 8 = 54). A tie error occurred only on tie quesrions when one tie number fact
was substituted for another (e.g., 8 x 8 = 81). The unknown errors did not fit any of the
above categories (e.g., 6 x 9 = 37). The NC Group made 11 multiplication errors overall
and the MD Group made 45 multiplication errors overall. Consistent with the literature,
more of the multiplication errors were table-related.

Table BI2

Totai Number of Each Type of Muiüplication Error


Group Table Close Tie Unknown
APPENDIX B

DISTANCE OF MJMBER FACT RESPONSE


FROM THE CORRECT ANSWER IN EACH OPERATION

For each error that was made on the Stage 2 number facts, the distance of the
response from the correct answer was calculated. For addition and subtraction, the
distance was measured in digits, so that an answer that was one digit away from the
correct answer would be identified as +/- 1. (e-g., 6 + 7 = 14). An answer that was two
digits away would be identified as +/- 2 (e-g., 6 + 7 = 15). Any responses further away
from the correct answer were identified as belonging to the +/-3 or more category.
For multiplication and division,o d y table-reiated enors were used for this
classification. Any emor that was one operand away from the correct answer (e-g., 6 x 7
= 48 is one operand away, as this is the correct answer for 6 x 8) was identified as +/- 1.
A response that was 2 operands away from the correct answer (e.g., 6 x 7 = 30 the 7 is
two operaods away, as this is the correct answer for 6 x 5) was identified as +/- 2. Table-
related errors that were 3 or more operands away from the correct answer were identified
as +/- 3 or more. In most cases the answers were +/- 1 mit away from the correct
answer.

Table BI3

Percentage of Respouses a Given Distance hthe Correct A m e r in Addition


Group +/- 1 +/12 +/- 3 or more

NC 66.6 16.7 16.7

MD 46.7 33.3 20.0


Table BI4

Percentage of Responses a Given Distance fimm the Correct A m e r Ui Snbtraction


Group +/- 1 +/12 +/- 3 or more

NC 37.0 48.2 14.8

Table BI5

Percentage of Responses a Given Distance th111 the C o m t Aaswer in Mnltiplication


Group 4- 1 +/1 2 +/- 3 or more

NC 87.5 O 12.5

MD 71.4 23.8 4.8

Table B16

Percentage of Responses a Given Distance h m the Correct m e r in Division


Group +/- 1 +/1 2 +/- 3 or more
NC 92.9 O 7.1

MD 72.0 28.0 O
SUBTYPE COMPARISONS

This appendix contains the charts and graphs associated with the analyses
comparing the NC,A and ARS Groups. It is organized in the same manner as the rest of
the thesis. For d l of these analyses, the estimated FSIQ score was used as a covariate.
The overall MANCOVA was sienificant (F [48,52] = 3.149, p < .ml),permitting further
investigation of the data. The Bonferroni criterion was set at p < .ûû2 for al1 contrasts
except when specifif tests were being analyzed further. In those cases separate alpha
levels were determined depending on the number of contrasts being made.

(i) Math Computation


A comparison of the 8 areas of the W R A J Arithrnetic subtest was carried out.
The results are presented in Table Cl. The Bonferroni criterion was set at p < .006 as
eight confmsts were being perfomed. A formal comparison of the errors was not done
due to the small sarnple sizes of the A and ARS Groups and the limited nurnber of errors.
Therefore, relevant error patterns will be discussed within the 8 area comparisons. Only
one graphornotor error was noted in d l of the responses and it was committed by a
student in group A, as Rourke (1993) would have predicteb However, the expected
outcome of more sign errors and more Failure to Shift Set errors in group A did not
materialize, Likely because the students in this group generally did not have the expected
pattern of visual-spatial deficits.
No simcant difference was found between the NC, A and ARS groups (F [2,
481 = 0.71,p > -10) on the addition questions. This finding is consistent with the NC
versus MD results. In contrast, no signif~canteffect was found between the NC, A and
ARS groups ( F [Z, 481 = 2.59, p > .05)for subtraction either. It was noted that the mean
scores of the A and ARS groups were both higher than the mean score of the MD group.
This suggests that it was the MD students who did not meet any of the subtyping criteria
that lowered the MD score and led to the slight difference found in the NC-MD conhast
Therefore, it appears that some students still have not caught up to the NC, but the A and
ARS students have. Most of the errors for both the A and ARS Groups were categorized
as sign or calculation errors. Only the Group ARS students made errors involving
subtractiog the larger number from the smdler number regardless of position.

Table Cl

WRAT3 Arithmetic Sobtest Item Analysis by Area for the NC, A and ARS Gmops

Number NC A ARS
of Mean Mean Mean
Questions (SD) (SD) (Sa
Addition 6 5.53 5.55 5.46
Cornputations (0.63) (0.69) (0.69)

Subtraction 5 4.70 4.00 436


Computations (0.60) (0.89) (0.67)

Multiplication 5 4.67 3.O9 2.91


Cornputations (0.61) (1.14) (0.94)
I

Division
Computations
1

Fractions 9 4.33 1.18 1.09


( 1 .58) (0.87) (0.94)

Percent

Algebral
Functions/
Conversions

** p < .O06 the significance Levei set with Bonferroni's Correction for this analysis.

A significant difference between the NC, A and ARS groups was found for both
multiplication (F [2,48] = 1880,p < .006) and division ( F [2' 481 = 1231, p < -006).
The difficulties the A and ARS Groups expenenced with these two operations were
similar to those noted in the NC - MD discussion, the 2 by 3-digit multiplication question
and the single-digit into multi-digit division question. The fraction questions were
padcularly difficult for the A and ARS Groups, with both of their mean scores falling
significantly below that of the NC Group (F [2,48] = 23.19, p -= .006).The questions
involving decimals and percent aiso differentiated between the NC. A and ARS Groups
(F (2,481 =12.65, p c .O06 and F [2,48] = 4.99, p < .05. respective1y). In the latter, the
NC Group did better than both the A and SARS Groups, but in the foxmer, the NC Group
only outperformed the ARS Group. No signifïcant differences were found for the
algebra, functions and conversion questions difficult (F [2,48] = 0.41, p r .IO)as there
was a floor effect on these questions. These cornputation findings indicate that overall
the A and ARS Groups perform similarly to each other. They also tend to score
signifkantly below the level of the NC Group in d areas except for addition, subtraction
and algebralfunctions/conversion

(ii) Basic Math Concepts


The findings from the basic math concepts cornparisons were very similar to those
found in the NC versus MD contrasts. The NC group scored ~ i ~ c a n thigher
l y than the
A and ARS groups on al1 five mesures of basic concepts (Numeration F [2,48] = 19.29,
Sequential and Quantitative Reasoning F [2,48] = 11.12, p < .002;Estimation
p c -002;
F [2,48] = 8.56, p < .ûû2 and Place Value F [2,48] = 330, p -= .O5). Therefore, both the
A and ARS subtypes appear to have an equivalent amount of trouble with these concepts.
The mean scores, standard deviations and statistical information for these anaiyses can be
found in Table C2.
Table C2

A Cornparison of NC,A and ARS Groups on Math Measpl~es

Math Measures A Post Hoc


Tukey HSD

47.27 NC > A, ARS


and Quantitative (7.48) (5.62)
Reasoning

Key Math - 1130 736 NC > A, ARS


Numeration ( 1.72) (136)

Estimation 3.O0 1.18 NC > A TARS


(max = 5 ) ( 1-02) (1-17)

Place Vaiue 2.97 2.09 NC > A, ARS


(max = 4) (-93 (.83

MANCOVA F [48,52] = 3.149, p < .ûû1


* p < .O5
** p < .ûûî the significance level set with Bonferroni's Correction for this anaiysis.

(;Ü) Namber Facts


A cornparisons between the number fact Mastery Rate (number correct/number of
questions) of the NC.A and ARS Groups was carried out for each of the operations. The
results are displayed in C 1. Univariate analyses of variance indicated that the NC
Group significantiy outperformed the NC, A and ARS groups, in al1 four operations
(addition F [2,48] = 16.19,p < -002,subtraction F [2,48] = 17.02, p c -002,
multiplication F [2,48] = 20.26, p i.O02 and division F [2,47 = 16.19, p -c .002). Post
hoc analyses iadicated that in addition the NC and A Groups were equivalent and both
were better than the ARS Group. In subtraction and division the NC Group was better
than the A Group, which was in tum better than the ARS Group. In multiplication, the
NC Group was better than both the A and ARS Groups. Thus in the area of number fact
speed, there do appear to be differences between the A and ARS Groups. The findings
here support the normal variation viewpoint.
Cornparisons between the three groups were also c h e d out for number fact
automaticity in each operation. Ia ail cases the NC group knew significantly more
number fact. automatically than the A and ARS Groups (addition F [2,48] = 14.56, p <
-002; subtraction F [2,48] = 7.83,
p < .ME; multiplication F [2,483 = 22.15,p c .02;
division F [2,48]= 23.52,p c -002). Figure C2 contains a graph of the rnean number of
questions answered automatically for each operation by each group. The limited sample
sizes in the A and ARS Groups and the small number of erron prevented a valid analysis
of number fact errors. The lack of diference between the A and ARS Groups for
automatic retrieval in addition suggests that Group A must perform their back-up
strategies very efficiently to be similar to the NC Group in the addition Mastery Rate.

(iv) Cognitive Measnres


Memory
It was only on the memory tests that sigdîcant differences between the A and
ARS groups were found. The results of the ANCOVA's comparing the NC,A and ARS
Groups indicated that there were significant differences among these three groups on the
Numbers Forward (F 12,481 = 3.59,p c .05), Sequences (F 12,481 = 1031, p < .002),
Reading Single-Digit (F [2,48] = 5.87,p c -05) and Two-Digit Numbers (F [2,48] =
12.95, p < ,002)tests and the Attention/Concentration composite (F [2,48] = 8-57,
p<
.002). Post hoc analyses indicated that the NC Group was ~ i ~ c a n tbetter
i y than the
ARS Group on the Numbers Forward and Reading Single- and Two-Digit Numbers tests
as well as the AttentiodConcentration composite. With the exception of Reading Single-
Digit Numbers,Group A also did better than Group ARS on these measures. The NC
Group did signifcantly betîer than both the A and ARS Groups on the Sequences
masure.
Table C3

Compariron Between the NC, A and ARS Groups on Memory Tests


Memory and NC A ARS ANCOVA ~ o sHoc
t
Attention Tests Mean Mean Mean (df -2,423) Tukey (HSD)
(SOI (Sm (SD) F P
CMS - Attention/ f 07.27 100.00 89-00 8.57 ml** NC > ARS
Concentration (9.26) (8.05) (2338) ABUS

C M S - Numbers 9.80 936 7.55 2.27 ns


Total (2.22) (2-58) (2.51)

CMS - Nurnbers 10.17 1036 7.64 3 -59 .03S NC > ARS


Fornard (2.02) (335) (3.01) A > ARS

CMS - Numbers 9.80 9.27 7.82 1.O5 ns


Backward (2.60) (2.10) (3-66)

CMS -Sequemes 12.60 10.64 8.9 1 1031 <.@Il**


( 1-92) ( 1.69) (2.26)

Read Single Digit 2.98 338 3.91 5.87 .005* NC < ARS
Numbers (0.63) (0.80) (0.98)

Read Double 5.08 5.08 7.54 12.95 c.ûûl** NC < ARS


Digit Numbers (1.18) (-70) (2.341 A < ARS

MANCOVA F 1423,521 = 3.149, p c .O01


* p c .O5
** p c .ûû2 the significance level set with Bonferroni's Correction for this andysis
Processinpr Speed
The univariate ANCOVA's yielded a signifïcant difference between the NC, A
and ARS groups for al1 of the processing speed m a s u r e s (Coding F [2,48]= 6.59, p <
-05;Symbol Search F [2,48] = 337, p < .05; Visual Matching F [2,48] = 12.66, p < ,002
and Planned Connections F [2,48] = 4.72,p c .CM2 and Processing Speed Factor score F
12,481 = 1230,p < -05).Post hoc analyses, indicated that the NC performed
significantly better than both the A and ARS groups in al1 instances. Once again,
however, the Symbol Search difference was not si@~cant at the more conservative p <
-003level. Although the trend appeared to be that group A scored higher than group
ARS, this pattern was only significant on the Visual Matching subtest. A review of the
Visual Matching protocols indicated that both groups made quivalent self corrections
(ARS = 14, A = 13) and that the Group A students made more errors (ARS = O, A = 5).
Therefore the differences between the two groups rnust lie in the scanning speed.

Table C4

Mean Seoros and Signincance Levels for A, ARS and NC Students on Professing
Speed Tests
Rocessing Speed 1
Tests

WSC-III -
-- -
NC A
Mean
(SD)
103.82
MGD)
?: 1 , ANCOVA
(df =2,48) 1 Post Hoc
Tukey

116.20
Processing Speed (12.65) (6-68)
WSC-III - 12.80 10.46
Coding (2.72) (1.81) (2.86)
WSC-III - 13.O3 10.64 9-64 337 -043" NC>A,ARS
Svrnboi Search (2.50) (2.77) (2.87)
85.91 12.66 -COOL** NC > A, ARS
Matching ( 12.62) (8.82) (1 1.12) A > ARS
CAS - Planned 12.73 10.18 9.55 4.72 .013* NC>A,ARS
Connections I (238) I (1.94)
MANCOVA F (48,521 = 3.149, p < .Ml1
* pc.05
** p c .ûû2 the sigaificance level set with Bonferroni's Correction for this andysis
Figure C l
Cornparison of Nnmber Fact Mastery Rate Between the NC,A and
ARS Gmups

Addition Subfraction Mu1ti plication Division


Operation
Figure C2
Cornparison of Number Fad Aotomaticity Between
the NC,A and ARS Groups

Addition Subtraction Multiplication Division


Operation

! uNC oA a ARS -I

You might also like