Está en la página 1de 19

690728

research-article2017
ECXXXX10.1177/0014402917690728Exceptional ChildrenPowell et al.

Article
Exceptional Children
2017, Vol. 83(4) 359377
Child-Level Predictors of The Author(s) 2017
DOI: 10.1177/0014402917690728
https://doi.org/10.1177/0014402917690728
Responsiveness to Evidence-Based journals.sagepub.com/home/ecx

Mathematics Intervention

Sarah R. Powell1, Paul T. Cirino2, and Amelia S. Malone3

Abstract
We identified child-level predictors of responsiveness to 2 types of mathematics intervention
(calculation and word problem) among second-grade children with mathematics difficulty.
Participants were 250 children in 107 classrooms in 23 schools pretested on mathematics and
general cognitive measures and posttested on mathematics measures. We randomly assigned
classrooms to calculation intervention, word-problem intervention, or business-as-usual control.
Intervention lasted 17 weeks. Path analyses indicated that scores on working memory and
language comprehension assessments moderated responsiveness to calculation intervention.
No moderators were identified for responsiveness to word-problem intervention. Across both
intervention groups and the control group, attentive behavior predicted both outcomes. Initial
calculation skill predicted the calculation outcome, and initial language comprehension predicted
word-problem outcomes. These results indicate that screening for calculation intervention should
include a focus on working memory, language comprehension, attentive behavior, and calculations.
Screening for word-problem intervention should focus on attentive behavior and word problems.

When a well-designed and competently exe- intervention depends on the childs pretest
cuted randomized controlled trial produces cognitive ability or academic skill. That is,
statistically significant effects favoring the children with low pretest performance respond
learning outcomes of children who receive less adequately than children with greater per-
a specified intervention over children who formance in the cognitive or academic domain.
do not receive it, the intervention is deemed One research team has investigated
evidence-based (Every Student Succeeds Act, whether domain-general and domain-specific
2015). This status suggests that most children variables moderated the effect of intervention.
responded well to the experimental interven- For example, in examining whether respon-
tion. Evidence-based status does not, however, siveness to a fraction intervention was associ-
mean that all children responded adequately ated with start-of-fourth-grade whole-number
(Cook & Cook, 2013). In fact, few if any inter- calculation skill, L. S. Fuchs, Sterba, Fuchs,
ventions achieve universal response, and little and Malone (2016) did not find a significant
is known about child characteristics associated
with inadequate response. This is especially
1
the case in mathematics, where few studies University of Texas at Austin
2
University of Houston
have examined child-level variables associ- 3
Vanderbilt University
ated with responsiveness in an at-risk sample.
To identify child-level response variables, the Corresponding Author:
Sarah R. Powell, Department of Special Education,
relevant moderation effect involves a prereq- University of Texas at Austin, 1 University Station
uisite cognitive ability or academic skill. With D5300, Austin, TX 78712.
a prerequisite response moderator, the effect of E-mail: srpowell@austin.utexas.edu
360 Exceptional Children 83(4)

moderator effect on posttest fraction under- the unknown. Along with this transparent dif-
standing or calculation performance. Children ference in the nature of these tasks, correla-
with a range of whole-number calculation tional studies find that a different set of cogni-
skill at pretest responded similarly to the tive abilities underlies calculation and
intervention. By contrast, L. S. Fuchs, Malone, word-problem skill in representative samples
etal. (2016) determined that responsiveness (e.g., Swanson, 2011). This suggests that the
to fraction word-problem intervention was child-level variables determining responsive-
associated with start-of-fourth-grade reason- ness to intervention may differ for these types
ing ability. That is, although most children of mathematics outcomes.
responded adequately, those with very low Our second extension to the literature con-
reasoning did not. cerned our target population: children with
Understanding how child-level variables concurrent low performance on calculations
influence childrens responsiveness to inter- and word problems. This focus is important
vention may help school personnel customize because children with concurrent difficulty
interventions to address at-risk childrens have more severe deficits in each domain than
needs. One implication of the work by L. S. do children with difficulty in one area (e.g.,
Fuchs, Malone and colleagues (2016) is that Cirino, Fuchs, Elias, Powell, & Schumacher,
children with very low reasoning ability 2015; L. S. Fuchs, Fuchs, Stuebing, etal.,
require a different form of, or a more intensive 2008; Willcutt etal., 2013). Concurrent diffi-
version of, fraction word-problem interven- culty therefore serves as an indicator of seri-
tion (e.g., more varied practice, hands-on ous risk for poor mathematics learning.
materials to establish fraction concepts, a self- Moreover, approximately twice as many chil-
regulation component for attention, more dren have concurrent difficulty than difficulty
intervention time). Such studies can also help in one area (L. S. Fuchs, Fuchs, Stuebing,
researchers identify directions for expanding etal., 2008). In this study, we refer to this
the overall efficacy of the intervention. population as children with mathematics dif-
ficulty (MD).
Understanding how child-level Our third extension was that our interven-
variables influence childrens tion relied on a multitier support system, in
responsiveness to intervention may which children with MD participated in
instruction that was aligned across a whole-
help school personnel customize
class intervention and a supplementary small-
interventions to address at-risk
group tutoring program. In providing aligned
childrens needs. instructional support, the hope was that small-
group tutoring worked synergistically with
whole-class mathematics instruction (L. S.
Extensions to the Literature
Fuchs, Fuchs, Craddock, etal., 2008; Lembke,
With the present study, we sought to extend Garman, Deno, & Stecker, 2010). This con-
prior work in four important ways. First, we text is important because a multitier frame-
contrasted child-level predictors of respon- work assumes that whole-class intervention
siveness to intervention on two forms of and small-group tutoring are strong. In the
mathematics intervention: whole-number cal- present study, this was the case for children
culations and word problems. Each area rep- with MD (Powell etal., 2015) and for the gen-
resents a critical form of mathematics eral population (L. S. Fuchs, Powell, etal.,
competence for success in school (National 2014). In the present study, we focused on a
Council of Teachers of Mathematics, 2014). A subset of children with MD who received both
calculation problem is set up for solution, instructional tiers.
whereas a word problem requires children to The final extension was methodological,
process text to build a problem model and then involving the use of path analysis. Modera-
construct a number sentence for calculating tion refers to a statistical interaction between
Powell et al. 361

(a) a child-level variable (i.e., cognitive abil- the effect between two conditions, the correla-
ity or mathematics skill) measured at the start tion between the child-level variable and the
of a study and (b) the effect between two study outcome is significantly stronger in one condi-
conditions on the studys academic outcome. tion than in the other. For this reason, the child-
With this interaction, the effect between two level variable is associated with the magnitude
study conditions on the outcome is larger or of the effect between the two conditions. This
smaller depending on pretest performance on is what we mean when we use the term mod-
an academic or cognitive variable. In most eration.
studies examining child-level variables asso- In the literature, predictors and moderators
ciated with response, researchers apply cut have been conceptualized in terms of domain-
points to childrens postintervention perfor- general abilities (e.g., working memory),
mance (or improvement over the course of which support performance across many aca-
intervention) to classify children as adequate demic areas, and domain-specific skills (e.g.,
versus inadequate responders. Then, research- understanding about discrete numbers), which
ers identify which child-level variables differ specifically support performance within an
significantly for the adequate and inadequate academic area. Gearys (2004) model of math-
responder groups. The drawbacks of this ematics learning indicated that mathematics
approach are that the selected cut point is arbi- performance depends on a constellation of
trary and that results vary with the stringency domain-general abilities and foundational
of the arbitrary cut point (e.g., D. Fuchs, mathematical skills. Our study reflects both
Compton, Fuchs, Bryant, & Davis, 2008; assumptions to understand how response vari-
Wise etal., 2008). Branum-Martin, Fletcher, ables related to cognition and academics may
and Stuebing (2012) therefore argued for an moderate responsiveness to intervention.
approach that treats performance continu- In the mathematics literature for children
ously rather than categorically. Path-analytic with MD, the following domain-general cog-
moderation analysis addresses Branum- nitive processes predicted individual differ-
Martin and colleagues concerns in two ways. ences in the development of procedural
By treating outcome performance continu- calculation and word-problem skill among
ously, it avoids discarding potentially infor- representative samples across environmental,
mative data on the full distribution of at-risk instructional, or intervention conditions. Rea-
performance. Second and relatedly, it empiri- soning (L. S. Fuchs, Geary, etal., 2010),
cally derives the cut point for responsiveness, working memory (e.g., Geary & Widaman,
as the score on the moderator variable at 1992), and attentive behavior (e.g., L. S.
which the effect of intervention transitions Fuchs etal., 2006) predicted competency with
from significance to nonsignificance. calculation and word-problem learning. Pro-
cessing speed predicted calculation learning
Rationale for the Potential (e.g., L. S. Fuchs, Geary, etal., 2010), and lan-
guage comprehension predicted word-
Moderators problem learning (e.g., Zheng, Swanson, &
As potential candidates for moderators, we Marcoulides, 2011). Given prior work, we
relied on the individual-differences literature to hypothesized main effects for reasoning,
identify child-level variables that predict devel- working memory, and attentive behavior on
opment in calculations or word problems. both outcomes; for processing speed on the
Some child-level variables predict develop- calculation outcome; and for language com-
ment in an academic area across environmen- prehension on the word-problem outcome. In
tal, instructional, or intervention conditions. In terms of domain-specific skills, we expected
this article, we use the term predictors to refer main effect predictors of end-of-second-grade
to variables that exert a main effect of a calculation learning to include pretest fluency
child-level variable on the outcome. By con- with facts and understanding about whole-
trast, when a child-level variable interacts with number magnitude (e.g., L. S. Fuchs etal.,
362 Exceptional Children 83(4)

2012). In the prediction of end-of-second-grade and domain-specific mathematics skills as


word-problem skill, we anticipated a main potential moderators. In isolating the effect of
effect predictive value for pretest word- each moderator (an interaction between the
problem skill as well as understanding about child-level variable and the effect of interven-
number (e.g., L. S. Fuchs etal., 2012). tion), we controlled for the effects of the other
seven variables on the outcome.
Extant Database for Data Analysis
We conducted our analyses on an extant data- Method
base (Powell etal., 2015) documenting the Participants
evidence-based status of a multilevel calculation
intervention and a multilevel word-problem We received Institutional Review Board
intervention for children with MD. In that approval and obtained signed consent forms
study, teachers were randomly assigned to cal- from teachers, consent forms from parents or
culation multitier intervention, word-problem guardians, and assent from children. With
multitier intervention, or a business-as-usual stratification by school, classroom teachers
condition. Multitier calculation intervention were randomly assigned to three conditions:
improved performance on the calculation out- ~37.5% to calculation intervention, ~37.5% to
comes, with an effect size (ES) of 1.67 on the word-problem intervention, and ~25% to con-
proximal outcome and 1.19 on the distal out- trol. Across four cohorts (Powell etal., 2015),
come. Multitier word-problem intervention we screened 1,917 children in 127 second-
enhanced performance on the proximal word- grade classrooms in 25 schools in a metropol-
problem outcome, with an ES of 1.31, but the itan school district using the following
effect on the distal outcome was not signifi- screening protocol to identify for MD (i.e.,
cant (ES = 0.15). In the current set of analyses, current difficulties with calculation and word
we indexed responsiveness at the end of sec- problems): Children had to score <7 on the
ond grade by creating a factor score across calculation screening measure (i.e., perfor-
proximal and distal measures of calculation mance on addition sums), <6 on the word-
performance and a factor score across proxi- problem screening measure (i.e., performance
mal and distal measures of word-problem per- on single-digit word problems), and >9th
formance. Note that an additional study (L. S. percentile on at least one subtest of the
Fuchs, Powell, etal., 2014) reported effects of two-subtest Wechsler Abbreviated Scale of
the interventions with a sample that included Intelligence (WASI; Wechsler, 1999). The
at-risk and not-at-risk children. benchmarks for the screeners were calculated
with data from a pilot study where we identi-
fied pretest cutoff scores predictive of
Summary of Purpose low performance on calculation and word-
The purpose of the present analyses was to problem outcomes. For this study, we adjusted
identify child-level predictors of responsive- these benchmarks downward so that we
ness to two types of mathematics intervention included only students <40th percentile in
among second-grade children with MD. Inter- both areas. This yielded 265 children from
vention involved an evidence-based supple- 110 second-grade classrooms in 25 schools.
mentary small-group instructional program Note that children were identified as MD in
that was aligned with an evidence-based the same way regardless of condition. These
classroom instructional program. To investi- children with MD constituted the lowest 14%
gate responsiveness, we applied moderation of the entire screened sample. Of these chil-
analyses to the extant database of at-risk dren, 96 were in the calculation intervention
children, for which prior analyses indicated (36%), 91 in the word-problem intervention
overall efficacy (L. S. Fuchs, Powell, etal., (34%), and 78 in the control group (29%). For
2014; Powell etal., 2015). We examined eight the present analyses, we eliminated 15 children
child-level domain-general cognitive abilities with incomplete data on the eight child-level
Powell et al. 363

variables investigated as potential moderators. ranged from .85 to .93. We used Sums to 12 as
The 250 children with complete data were in the calculation screening measure for deter-
107 classrooms in 23 schools, and 91 were in mining MD status.
the calculation intervention (36%), 90 in the Story Problems (Jordan & Hanich, 2000)
word-problem intervention (36%), and 69 in comprised 14 problems representing com-
the control group (28%). Table 1 features bine, compare, and change word-problem
demographic information and mathematics types that involve single-digit addition or
performance by condition. subtraction for solution. The test was untimed.
The examiner read each word problem aloud
as the children followed along on paper; then,
Screening Measures the examiner provided time for children to
We used three measures to screen for MD sta- write answers. A childs score was the number
tus. At pretest, we administered four subtests of correct answers. Cronbachs was .87.
of the Second-Grade Calculations Battery The WASI (Wechsler, 1999) is an individu-
(SGCB; L. S. Fuchs, Hamlett, & Powell, ally administered measure of general cogni-
2003): Sums to 12, Sums to 18, Minuends to tive ability with two subtests. Vocabulary
12, and Minuends to 18. All four subtests measured word knowledge and expressive
required single-digit addition or subtraction. language with 4 picture items and 37 vocabu-
For each subtest, children had 1 min to lary words. Children defined words until the
complete up to 25 problems. Score was the end of the test or until reaching a ceiling of
number of correct answers. Cronbachs five consecutive errors. Split-half reliability

Table 1. Demographics and Mathematics Performance by Condition.

Calculation Word-problem
intervention intervention Control

Variable % M SD % M SD % M SD
Demographics
Male 49 60 41
Subsidized lunch 93 87 90
Race
African American 55 55 58
White non-Hispanic 11 16 19
White Hispanic 30 22 17
Other 4 7 6
Special education 9 9 14
English learner 18 17 16
Pretest mathematics
Number line 17.97 6.04 19.25 6.37 18.62 6.35
Single-digit fluency 11.53 5.08 11.58 5.10 11.99 5.17
Story problems 2.97 1.55 3.21 1.58 3.12 1.59
Posttest mathematics
Single-digit fluency 44.54 16.38 29.84 11.66 21.97 9.83
Double-digit accuracy 21.92 8.98 13.04 7.01 10.03 6.71
KeyMath-calculations 16.42 4.03 13.53 3.04 12.33 3.53
Second-grade word problems 2.87 1.83 5.70 3.21 3.35 1.91
KeyMathproblem solving 2.26 1.44 2.40 1.24 2.28 1.40
ITBSdata interpretation and problem solving 13.48 3.79 13.17 3.40 12.90 3.23

Note. ITBS = Iowa Test of Basic Skills.


364 Exceptional Children 83(4)

is .86. Matrix Reasoning measured nonverbal numbers. They had 3 min to complete 60
reasoning with pattern completion, classifica- rows. Per the test developer, reliability is .91.
tion, analogy, and serial reasoning tasks. Chil-
dren chose the best of five choices to complete Reasoning. To index initial reasoning, we used
a visual pattern for 35 items or until reaching the Matrix Reasoning subtest of the WASI
a ceiling of four consecutive errors or four (Wechsler, 1999), as already described.
errors over five consecutive items. Reliability,
as presented in the manual, is .94. The two Working memory. We used two dual-task cen-
subtest scores combine to yield an Estimated tral executive subtests from the Working
Full Scale IQ score with reliability of .92. Memory Test Battery for Children (Pickering
& Gathercole, 2001). Each had six items at
span levels from 16 to 19. Passing four
Moderator Measures
items at a level moved the child to the next
Addition and subtraction. To index initial cal- level. At each span level, the number of items
culation skill, we used two subtests of the to be remembered increased by one. Failing
SGCB (L. S. Fuchs etal., 2003). Sums to 12 three items terminated the subtest. For Listen-
comprised 25 addition fact problems with ing Recall (referred to here as working memory
sums from 6 to 12. Children had 1 min to sentences), the child determined if each
write answers. Minuends to 12 comprised 25 sentence in a series was true and then recalled
subtraction fact problems with minuends from the last word in each sentence. For Counting
5 to 12. Children write differences for 1 min. Recall (referred to here as working memory
Cronbachs was .95. counting), the child counted a set of 4 to 7
dots, each on a separate card. After the last
Story Problems. To index initial word-problem card, the child recalled the number of dots on
skill, we used Story Problems (Jordan & each card. Per the test developer, test-retest
Hanich, 2000), as already described. reliability is .91 to .93. We opted to include
both subtests, rather than create a composite
Attentive behavior. SWAN (Strengths and Weak- variable, based on prior work suggesting indi-
nesses of ADHD-symptoms and Normal behav- vidual differences in working memory for
ior) is an 18-item teacher rating scale (Swanson numbers versus words (Raghubar, Barnes, &
etal., n.d.) that sampled items from criteria Hecht, 2010).
of the Diagnostic and Statistical Manual of
Mental Disorders (fourth edition; American Language comprehension. The Woodcock Diag-
Psychiatric Association, 1994) for attention- nostic Reading Battery Listening Comprehension
deficit/hyperactivity disorder for inattention (Woodcock, 1997) measures the ability to
(Items 19) and hyperactivity or impulsivity understand sentences or passages. With 38
(Items 1018). Items were rated from 1 (far items, children supplied the word missing
below) to 7 (far above). We report data for the at the end of sentences or passages that
Attentive Behavior subscale as the average rat- progressed from simple verbal analogies and
ing across the nine relevant items. We selected associations to discerning implications. The
this subscale to index attentive behavior, or the test manual provided examples of correct
ability to maintain focus of attention. The responses to guide scoring. Reliability, as
SWAN correlates well with other dimensional reported by the test developer, is .80.
assessments of behavior related to attention.
Cronbachs in this study was .98. Number line.With Number Line Estimation
(Siegler & Booth, 2004), children located
Processing speed. With Woodcock-Johnson III Arabic numerals on a horizontal number line
Visual Matching (Woodcock, McGrew, & marked with 0 and 100 as end points. A num-
Mather, 2001), children located and circled ber line was presented with a target number
two identical numbers in each row of six shown above the line. Children placed the
Powell et al. 365

target number on the number line, without a decimals, and negative numbers. We admin-
time limit. The examiner explained, The istered both KM subtests individually.
number 50 is half of 100, so we put it halfway in Cronbachs for the two subtests was .84
between 0 and 100 on the number line. Use of and .81, respectively. Analysis indicated that
these directions produced findings comparable a single-factor solution across these seven
to other work (e.g., Geary, Hoard, Byrd-Craven, tests (three SGCB subtests, Two-Digit
Nugent, & Numtee, 2007). Siegler and Opfer Addition and Two-Digit Subtraction, and
(2003) used group-level median placements KM Addition and KM Subtraction) was a
fitted to linear and log models to make infer- good fit to the posttest data.
ences about childrens modal representation
for making placements but used an accuracy Word problems.We used three measures to
measure for individual difference analyses. index word-problem performance. Second-
The accuracy measure was a good indicator of Grade Word Problems (L. S. Fuchs, Zumeta,
overall understanding of the number line and etal., 2010), the proximal measure, included
predicted later mathematics achievement 18 word problems representing combine,
(Geary, 2011). Lower scores indicated stron- compare, and change problems types. Solu-
ger performance, but we multiplied scores by tions required one-digit addition or subtrac-
1 before conducting analyses. Test-retest tion. In groups, the examiner read each
reliability at second grade has been reported at problem aloud as children followed along on
.87 (L. S. Fuchs, Geary, Fuchs, Compton, & paper. Children had 1 min to write a con-
Hamlett, 2014). structed response. Children earned credit for
correct word-problem calculation and label to
reflect their processing of the problem state-
Outcome Measures ment and understanding of the problems
Calculations. We used three measures. First, focus. Cronbachs was .88. We measured
we indexed single-digit calculation skill distal effects with KM Problem Solving
with the three nonscreening subtests of the (Connolly, 1998) and the Iowa Test of Basic
SGCB (L. S. Fuchs etal., 2003): Sums to 18, SkillsData Interpretation and Problem
Minuends to 12, and Minuends to 18. The Solving (Hoover, Hieronymous, Dunbar, &
outcome was the average of these three sub- Frisbie, 1993). KM Problem Solving included
tests. Second, to measure proximal double- 18 word problems of increasing difficulty,
digit calculation, we administered Two-Digit which involved all four operations on word
Addition and Two-Digit Subtraction of the problems with transparent solution strategies,
SGCB (L. S. Fuchs etal., 2003). For each nonroutine problems without clear solution
subtest, children had 3 min to complete up to strategies, and items requiring children to
20 problems in which two-digit numbers demonstrate comprehension of a word prob-
were added or subtracted. Cronbachs lem without solving it. Administration was
for Two-Digit Addition and Two-Digit individual, and items were presented verbally
Subtraction was .96 and .87, respectively. Third, with accompanying pictures. Children
we measured distal effects with KeyMath- answered with a verbal response. Testing was
Revised (KM; Connolly, 1998) Addition and discontinued after three consecutive errors.
Subtraction. On KM Addition, children The score was the number of correct multiple-
answered six problems with pictures and 12 choice answers. Cronbachs was .74. The
problems in a written format. Problems Iowa Test of Basic Skills was administered in
ranged from one-digit addition to addition groups. Children worked on 8 word problems
with decimals, fractions, and negative numbers. presented verbally and 22 written problems.
KM Subtraction functioned similarly with Problems included one- and two-digit calcula-
six problems presented with graphics and 12 tion with and without irrelevant information,
written problems of one-, two-, and three-digit with and without charts or graphs, and with
subtraction with whole numbers, fractions, and without multiple steps. The examiner read
366 Exceptional Children 83(4)

problems, and children responded using a subtraction facts (i.e., single- or double digit
multiple-choice format. Cronbachs was minuend and a single-digit subtrahend). Prac-
.81. Analysis indicated that a single-factor tice helped build fluency, which in turn
solution across measures represented a good helped improve double-digit addition and
fit to the posttest data. subtraction. RAs also explicitly taught
efficient procedures for identifying when
regrouping was required in addition and
Interventions
subtraction problems and how regrouping
Children participated in multitiered calcula- differed in addition and subtraction.
tion intervention, multitiered word-problem
intervention, or business-as-usual control. Word-problem framework. Word-problem inter-
The general education instructional tier com- vention incorporated Kintsch and colleagues
prised 34 whole-class lessons (2 lessons per framework in which word-problem solving is
week over a 17-week period). Each lesson thought to engage problem solving and lan-
lasted 40 to 45 min. Research assistants guage comprehension (Cummins, Kintsch,
(RAs) delivered all lessons in the childrens Reusser, & Weimer, 1988; Kintsch & Greeno,
classrooms, and classrooms included 15 to 1985). This model of word-problem solving
26 students. Small-group intervention com- suggests that general features of text apply
prised 39 lessons. Tutoring started at the start across all word problems but that word prob-
of Weeks 4 or 5 of the general education lems differ by problem type. Children learn to
series of lessons and lasted 13 weeks (3 les- apply a schema to assist with solving specific
sons per week). Another team of RAs pro- word problems, and at second grade, the three
vided small-group intervention in groups of 2 whole-number additive schemas include com-
to 3 children, and each lesson lasted 25 to 30 bine, compare, and change (Riley & Greeno,
min. Small-group intervention occurred out- 1988).
side the childs classroom and did not occur
at the same time in the school day as the gen- Fidelity. To assess fidelity, all sessions were
eral education instructional tier. That is, chil- digitally audio-recorded, and we conducted
dren received the combination of whole-class in-school observations of each RA deliver-
instruction plus small-group intervention. All ing intervention. Prior to the first session,
RAs were working on or had finished gradu- we created checklists of essential informa-
ate degrees in education-related fields and tion for each lesson within the calculation
received extensive training before working intervention and the word-problem interven-
with children. tion. At the end of each school year, RAs
independently listened to a sample of record-
Calculation framework. Whole-class and small- ings selected at random while completing
group calculation intervention incorporated checklists to identify the percentage of
two areas of emphasis for developing concep- points addressed. For small-group tutoring,
tual and procedural competence with one- and the mean percentage of points addressed
two-digit addition and subtraction problems exceeded 95 in each condition.
(e.g., L. S. Fuchs etal., 2013; Groen &
Resnick, 1977). We focused on interconnected Additional information. For additional informa-
knowledge about number and number fami- tion on teachers (nonevidence based) class-
lies. Children used number families to explore room instruction, the structure of the
the inverse relationship between addition and calculation and word-problem whole-class
subtraction. We also included instruction on intervention, the structure of the calculation
place value concepts of tens and ones. The and word-problem small-group intervention,
other area for emphasis was practice. Children the RAs, their training, and fidelity, see L. S.
learned counting strategies for solving addi- Fuchs, Powell, etal. (2014) and Powell etal.
tion facts (i.e., two single-digit addends) and (2015).
Powell et al. 367

Procedure Eight pretest mathematics skills and cogni-


tive abilities were evaluated as potential mod-
We administered the screening measures in erators (i.e., determinants of childrens
September and the pretest measures in October. responsiveness to intervention): two pretest
RAs delivered whole-class intervention from mathematics skills (single-digit addition and
November to March, and a separate team of subtraction fluency or story problems; num-
RAs delivered small-group intervention from ber line understanding) and six pretest cogni-
December to March. We posttested children tive abilities (reasoning, attentive behavior,
in late March. On individual measures, we processing speed, working memory
audio-recorded each administration and sam- sentences, working memorycounting, and
pled 20% of tapes for rescoring. Agreement language comprehension). In each analysis,
exceeded 99%. RAs independently entered one of the eight pretest variables was tested as
responses on 100% of the test protocols for the moderator while controlling for the other
each measure on an item-by-item basis into seven pretest variables. Each moderation
two separate databases. We rectified the analysis was a test of a statistical interaction
discrepancies between the two databases to between (a) the intervention effect (interven-
reflect the childs original response. tion vs. control group) and (b) the pretest abil-
ity or skill (treated continuously). A significant
interaction or moderation indicated that the
Data Analysis
level of the childs performance on the pretest
Prior to conducting analyses, we transformed variable was associated with responsiveness
scores on each measure to sample-based z to intervention.
scores. The moderation analyses tested To follow up each significant interaction,
whether a pretest skill or ability interacted we relied on the Johnson-Neyman technique
with the main effect, contrasting the interven- (Bauer & Curran, 2005; Hayes & Matthes,
tion against the control group (thereby deter- 2009), which derived the value along the
mining childrens responsiveness to the full continuum of the pretest ability or skill
intervention). Following Preacher and Hayes at which the effect of intervention transi-
(2008; Hayes, 2013), we used an ordinary tioned from statistically significant to non-
least squares path-analytic framework. Mod- significant. In this way, the regions of
eration analysis, based on a path-analytic significance for the intervention effect were
framework, has been used to investigate empirically derived and corresponded to
responsiveness to intervention (e.g., L. S. values of the pretest ability or skill at which
Fuchs, Malone, etal., 2016). The Preacher children demonstrated adequate response to
and Hayes analyses permitted (a) conclusions the intervention: children with pretest scores
about which pretest child-level variables exceeding the value (empirically derived cut
interact with the main effects of intervention point) at which the effect of intervention
and (b) empirical information about where transitioned to significance. The regions of
along the distribution of that pretest ability or nonsignificance corresponded to values of
skill the effect of intervention transitions from the pretest ability or skill at which children
significant to nonsignificant. (Note that did not benefit from the intervention: chil-
Barrett [2007] suggested a minimum sample dren with pretest scores falling below the
size of 200, whereas Kline [2016] indicated a value (empirically derived cut point) at
minimum of 10 subjects per parameter. which the effect of intervention transitioned
Although these analyses did not meet the Bar- to nonsignificance. Each significant moder-
rett criterion, they did meet the Kline guide- ation (interaction) effect demonstrated that a
line [i.e., 11 parameters for each model, with pretest ability or skill was linked to chil-
sample sizes of 159 and 160]. Moreover, drens responsiveness to intervention and
allowances were made for populations that provided the cut point on the pretest variable
are restricted in size, as in this case.) associated with responsiveness.
368 Exceptional Children 83(4)

This approach did not require arbitrarily The second moderator of the main effect
set cut scores for designating responsiveness. favoring calculation intervention over the
Instead, the Johnson-Neyman analysis empir- control group involved language comprehen-
ically derived that point by testing the differ- sion. For the model, R2 was .62, F(10, 149) =
ence between intervention and control groups 23.82, p < .001. The path coefficient for the
for every value of the pretest ability or skill. interaction was 0.22 (SE = 0.11), t = 2.11, p =
For each significant moderator, we provide a .04. We again relied on the Johnson-Neyman
figure to help visualize the nature of the mod- technique to probe this interaction. Figure 2 is
erator (interaction or responsiveness) effect. a visualization of this interaction. The x-axis
shows illustrative points along the distribution
of language comprehension percentile ranks
Results for the present sample. As shown, the effect of
Moderators of Responsiveness to the calculation intervention was smaller for
Calculation Intervention children with lower language comprehension
as compared with children with more ade-
Table 2 shows the raw and standard score quate language comprehension: The differ-
means, SDs, and correlations for variables ence between the black bars (intervention
included in the analyses. Table 3 features group) and the white bars (control group)
results of the responsiveness (moderation) decreased as children language comprehen-
analyses. We identified two significant mod- sion decreased. The effect favoring the calcu-
erator effects. Working memorysentences lation intervention over the control group
moderated the main effect between the calcu- transitioned from significant to nonsignificant
lation intervention and the control group on at 2.79 SDs below the sample mean on lan-
the calculation outcome. For this model, R2 guage comprehension.
was .62, F(10, 149) = 23.86, p < .001. The
path coefficient for the interaction was 0.22
(SE = 0.10), t = 2.15, p = .03. To probe this Moderators of Responsiveness to
interaction between working memory Word-Problem Intervention
sentences and the effect of the calculation
We ran eight analogous models contrasting
intervention over the control group, we relied
the word-problem intervention group against
on the Johnson-Neyman technique (Bauer &
the control group on the word-problem factor
Curran, 2005; Hayes & Matthes, 2009). This
outcome score but substituting pretest word-
technique derives the value, along the full
problem skill (i.e., Story Problems) for single-
continuum of the moderator, at which the
digit fluency (see Tables 2 and 3). We
effect of X on Y transitions from statistical sig-
identified no moderators of the main effect
nificance to nonsignificance. Figure 1, top
favoring the word-problem intervention over
panel, is a visualization of this interaction.
the control group.
The x-axis shows illustrative points along the
distribution of working memory percentile
ranks for the present sample. As shown, the Predictors of Calculation and Word-
effect of the calculation intervention was
smaller for children with lower working
Problem Learning Across Groups
memory as compared with children with more Across intervention and control groups,
adequate working memory. That is, the differ- start-of-second-grade teacher rating of atten-
ence between the black bars (the calculation tive behavior was a significant predictor of
intervention groups scores) and white bars both mathematics outcomes (see Table 3).
(the control groups scores) decreased as chil- Also, initial calculation skill was a signifi-
drens working memory capacity decreased. cant predictor of the calculation outcome,
Even so, the effect favoring the calculation and pretest language comprehension was a
intervention over the control group remained significant predictor of the word-problem
significant. outcome.
Powell et al. 369

Table 2.Means, SDs, and Correlations for Intervention and Control Groups.
Raw score Standard score Correlations

Variables M SD M SD AS/SPa A P R W-S W-C L N

Calculation intervention and control groups (n = 160)


Pretest add and subtract (AS) 11.73 5.53 NA
Attentive behavior (A) 31.45 11.35 NA .21
Processing speed (P) 10.33 2.28 90.28 11.99 .14 .12
Reasoning (R) 12.82 5.49 42.69 7.05 .25 .29 .20
Working memory 5.49 3.58 76.98 16.73 .12 .36 .18 .42
sentences (W-S)
Working memorycounting 12.71 3.91 79.38 13.66 .11 .16 .07 .24 .26
(W-C)
Language comprehension (L) 14.76 4.30 82.02 11.29 .14 .16 .08 .07 .30 .11
Number line (N) 18.27 6.33 NA .32 .18 .07 .10 .21 .12 .05
Posttest calculation skillb 0.00 1.00 NA .31 .40 .16 .25 .24 .21 .05 .22
Word-problem intervention and control groups (n = 159)
Pretest story problems (SP) 3.17 1.56 NA
Attentive behavior (A) 29.88 10.33 NA .09
Processing speed (P) 10.25 2.26 89.49 11.91 .03 .20
Reasoning (R) 12.50 4.73 42.79 6.98 .23 .25 .19
Working memory 5.24 3.66 77.26 19.29 .22 .29 .13 .31
sentences (W-S)
Working memorycounting 12.53 3.69 78.27 12.41 .15 .20 .04 .23 .18
(W-C)
Language comprehension (L) 15.33 4.67 83.92 12.99 .10 .04 .07 .11 .33 .05
Number line (N) 19.00 6.51 NA .18 .08 .08 .04 .19 .10 .13
Posttest word problems 0.00 1.00 NA .18 .41 .16 .22 .27 .13 .27 .17

Note. Pretest add and subtract on Second-Grade Calculations Battery (L. S. Fuchs, Hamlett, & Powell, 2003);
attentive behavior from Strengths and Weaknesses of ADHDSymptoms and Normal Behavior (Swanson et al., n.d.);
processing speed on Visual Matching from Woodcock-Johnson III (Woodcock, McGrew, & Mather, 2001); reasoning
on Matrix Reasoning from Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999); working memorysentences
is Listening Recall from the Working Memory Test BatteryChildren (Pickering & Gathercole, 2001); working
memorycounting is Counting Recall from the Working Memory Test BatteryChildren; Listening Comprehension
from Woodcock Diagnostic Reading Battery (Woodcock, 1997); Number Line Estimation (Siegler & Booth, 2004);
NA = not applicable.
a
AS for Calculation intervention and control groups (n = 160); SP for Word-problem intervention and control groups
(n = 159). bPosttest calculation is a factor score across single-digit addition and subtraction fluency (L. S. Fuchs et al.,
2003), double-digit calculation skill (L. S. Fuchs et al., 2003), and KeyMath addition and subtraction (Connolly, 1998).

Discussion (L. S. Fuchs etal., 2012; Vukovic etal.,


2014), yet little is known about potential
With the present set of analyses, our goal was moderators of success in these areas
to deepen insight into the child-level vari- among second-grade children with MD. We
ables that contribute to inadequate response investigated eight potential moderators of
to evidence-based mathematics intervention. intervention.
Few studies have addressed this issue in Second, our target population included
mathematics, and the present set of analyses children with concurrently low calculation
extends prior work in four ways. First, we and word-problem performance. These chil-
investigated potential domain-general and dren are especially at risk for more severe
domain-specific moderators of responsiveness mathematics deficits (e.g., L. S. Fuchs, Fuchs,
to intervention for two contrasting conditions Stuebing, etal., 2008) and tend to experience
calculations and word problems. Proficiency poorer outcomes on higher-order mathematics
with calculations and word problems is foun- than children with deficits in just one area of
dational for future mathematics success mathematics. Identifying potential moderators
Table 3. Moderation Analyses for Intervention and Control Groups.

370
Add and subtract / Working memory Working memory Language
Story problemsa Attentive behavior Processing speed Reasoning sentences counting comprehension Number line

Effects Coeff t(p) Coeff t(p) Coeff t(p) Coeff t(p) Coeff t(p) Coeff t(p) Coeff t(p) Coeff t(p)

Moderators for calculation intervention and control groups (n = 160)


Constant 0.01 0.01 (.990) 0.01 0.22 (.826) 0.00 0.03 (.978) 0.00 0.07 (.947) 0.00 0.02 (.986) 0.00 0.01 (.966) 0.00 0.06 (.950) 0.00 0.01 (.991)
Moderator 0.26 4.44 (<.001) 0.15 2.59 (.010) 0.04 0.76 (.450) 0.03 0.43 (.665) 0.11 1.81 (.073) 0.10 1.81 (.072) 0.01 1.18 (.239) 0.04 0.77 (.441)
Intervention 1.24 11.54 (<.001) 1.24 11.63 (<.001) 1.24 11.55 (<.001) 1.24 11.58 (<.001) 1.24 11.70 (<.001) 1.24 11.52 (<.001) 1.24 11.72 (<.001) 1.24 11.53 (<.001)
Moderator Intervention 0.03 0.31 (.761) 0.12 1.12 (.266) 0.06 0.55 (.586) 0.12 1.12 (.245) 0.22 2.15 (.033) 0.03 0.24 (.813) 0.22 2.11 (.036) 0.02 0.21 (.836)
Control variables
Add and subtract NA 0.26 4.52 (<.001) 0.26 4.42 (<.001) 0.26 4.49 (<.001) 0.26 4.52 (<.001) 0.26 4.43 (<.001) 0.26 4.56 (<.001) 0.26 4.44 (<.001)
Attentive behavior 0.15 2.61 (.010) NA 0.16 2.70 (.008) 0.16 2.75 (.007) 0.16 2.85 (.005) 0.16 2.67 (.008) 0.16 2.72 (.007) 0.15 2.64 (.009)
Processing speed 0.05 0.88 (.381) 0.05 0.91 (.364) NA 0.05 0.90 (.369) 0.05 0.95 (.346) 0.05 0.85 (.396) 0.04 0.79 (.431) 0.05 0.86 (.389)
Reasoning 0.03 0.42 (.672) 0.02 0.39 (.694) 0.03 0.46 (.644) NA 0.01 0.15 (.882) 0.03 0.46 (.647) 0.01 0.11 (.911) 0.03 0.46 (.643)
Working memory 0.09 1.42 (.157) 0.10 1.52 (.131) 0.09 1.45 (.148) 0.09 1.39 (.167) NA 0.09 1.41 (.161) 0.11 1.70 (.091) 0.09 1.41 (.161)
sentences
Working memory 0.10 1.80 (.073) 0.11 1.97 (.051) 0.10 1.78 (.077) 0.11 1.98 (.049) 0.10 1.84 (.064) NA 0.95 1.78 (.078) 0.10 1.81 (.072)
counting
Language comprehension 0.04 0.67 (.504) 0.04 0.70 (.484) 0.03 0.63 (.532) 0.03 0.55 (.582) 0.04 0.87 (.386) 0.04 0.65 (.515) NA 0.04 0.64 (.525)
Number line 0.04 0.81 (.420) 0.05 0.83 (.500) 0.04 0.78 (.436) 0.03 0.61 (.544) 0.04 0.77 (.442) 0.04 0.80 (.428) 0.05 0.95 (.342) NA
Moderators for word-problem intervention and control groups (n = 159)
Constant 0.00 0.03 (.978) 0.00 0.06 (.952) 0.00 0.01 (.989) 0.00 0.00 (.999) 0.00 0.00 (.999) 0.00 0.02 (.988) 0.00 0.01 (.994) 0.00 0.04 (.965)
Moderator 0.07 1.03 (.303) 0.34 4.53 (<.001) 0.06 0.80 (.422) 0.07 0.97 (.331) 0.02 0.25 (.800) 0.02 0.27 (.787) 0.20 2.78 (.006) 0.11 1.60 (.111)
Intervention 0.41 3.00 (.003) 0.40 3.03 (.003) 0.42 3.03 (.003) 0.42 3.03 (.003) 0.42 3.02 (.003) 0.42 3.01 (.003) 0.42 3.02 (.002) 0.42 3.06 (.003)
Moderator Intervention 0.13 0.90 (.371) 0.10 0.72 (.470) 0.11 0.78 (.436) 0.08 0.54 (.588) 0.00 0.02 (.982) 0.10 0.72 (.475) 0.01 0.10 (.924) 0.14 0.98 (.328)
Control variables
Story problems NA 0.07 0.91 (.364) 0.08 1.04 (.302) 0.07 0.96 (.338) 0.08 1.04 (.302) 0.07 1.01 (.316) 0.07 1.01 (.315) 0.07 0.98 (.327)
Attentive behavior 0.34 4.54 (<.001) NA 0.33 4.41 (<.001) 0.34 4.51 (<.001) 0.33 4.41 (<.001) 0.34 4.54 (<.001) 0.33 4.40 (<.001) 0.32 4.27 (<.001)
Processing speed 0.06 0.88 (.382) 0.06 0.90 (.370) NA 0.07 0.96 (.337) 0.05 0.95 (.346) 0.06 0.89 (.374) 0.07 0.94 (.349) 0.07 0.96 (.338)
Reasoning 0.08 1.07 (.287) 0.08 1.02 (.310) 0.08 1.03 (.303) NA 0.08 0.99 (.323) 0.08 1.06 (.289) 0.05 0.69 (.494) 0.08 1.07 (.286)
Working memory 0.03 0.27 (.784) 0.03 0.35 (.726) 0.03 0.36 (.722) 0.03 0.31 (.760) NA 0.03 0.29 (.770) 0.07 0.96 (.339) 0.02 0.26 (.794)
sentences
Working memory 0.01 0.17 (.867) 0.02 0.31 (.753) 0.01 0.19 (.853) 0.02 0.30 (.763) 0.02 0.21 (.836) NA 0.03 0.29 (.791) 0.02 0.21 (.834)
counting
Language comprehension 0.21 2.91 (.004) 0.21 2.87 (.005) 0.20 2.75 (.007) 0.20 2.69 (.008) 0.20 2.78 (.006) 0.21 2.82 (.005) NA 0.29 2.78 (.006)
Number line 0.11 1.53 (.128) 0.12 1.69 (.093) 0.11 1.60 (.112) 0.11 1.60 (.112) 0.12 1.61 (.109) 0.12 1.62 (.108) 0.12 1.61 (.109) NA

Note. Coeff = coefficient; NA = not applicable. aAdd and subtract for Moderators for calculation intervention and control groups (n = 160); Story problems for Moderators for word-problem intervention
and control groups (n = 159).
Powell et al. 371

of intervention for these groups of children Intervention outcomes were stronger for chil-
has important implications for designing dren with greater working memory capacity
future effective interventions. than for those with less working memory
Third, both active treatment conditions capacity. With this correlation in the interven-
involved a multitier support system, in which tion condition, the ES favoring the calculation
children with MD participated in instruction intervention over the control group decreased
that was aligned across the general education as working memory decreased. At the 90th
classroom and the supplementary small-group percentile of the samples distribution of
intervention program. That is, children in the working memory, the ES was 1.45; at the 10th
calculation or word-problem condition received percentile, 0.89. Although the effect favoring
whole-class instruction combined with small- the calculation intervention over the control
group intervention. This is a novel design, as group remained statistically significant within
most responsiveness-to-intervention random- the observable range of this moderator, this
ized controlled trials do not include two pattern along with the significance of the
evidence-based tiers of support for children interaction effect indicates that working mem-
with MD, even though multitier systems of ory is associated with responsiveness to inter-
support specify two strong instructional tiers. vention. This finding illustrates the importance
The fourth extension was relying on path- of moderator analyses even for generally effi-
analytic moderation analysis, which treats cacious interventions. Although effects were
child-level variables continuously and empiri- strong throughout the range of performance,
cally derives cut points for responsiveness. findings suggest that schools consider addi-
When a child-level variable interacts with an tional strategies to address the needs of chil-
effect between an intervention and control dren with extremely limited working memory
condition, the correlation between the child- capacity, just as results indicate that research-
level variable and the outcome is stronger in ers might further strengthen the intervention
one condition than the other, and the child- to address the needs of all learners.
level variable is associated with the magni-
tude of the effect between the two conditions. Intervention outcomes were stronger
We considered eight cognitive abilities and for children with greater working
mathematics skills as potential moderators, memory capacity than for those with
and in each analysis, we controlled for the less working memory capacity.
effect of the other seven.
For analyses contrasting the calculation The nature of the second moderator effect,
intervention against the control condition on involving language comprehension, was sim-
the calculation outcome, we identified two sig- ilar. As visualized in Figure 2, the control
nificant moderator effects. The first was work- group did similarly poorly, regardless of chil-
ing memory. As visualized in the top panel of drens language comprehension ability, with
Figure 1, the control group did similarly a nonsignificant correlation between lan-
poorly, regardless of childrens working mem- guage comprehension and the outcome score.
ory capacity; that is, the correlation between In the calculation condition, outcome scores
working memory and outcome score was not were substantially better than for the control
significant in the control group. The condition group throughout the range of initial lan-
in which children received the multilevel cal- guage comprehension performance, but the
culation intervention performed substantially correlation between language comprehension
better than the control group. This was true and posttest calculation score was significant:
throughout the range of initial working mem- Intervention outcomes were stronger for chil-
ory scores. dren with more intact language comprehen-
Yet, in the calculation intervention, the cor- sion than children with weaker language
relation between working memory and the comprehension. The ES favoring the calcula-
posttest calculation score was significant: tion condition over the control group
372 Exceptional Children 83(4)

Figure 1. Interaction between working memorysentences and the effect of calculation intervention
over the control group.

Figure 2. Interaction between language comprehension and the effect of calculation intervention over
the control group.
Powell et al. 373

decreased as the samples language compre- during counting up. With double-digit addi-
hension scores decreased. At the 90th percen- tion and subtraction, children must utilize
tile, the ES was 1.51; at the 10th percentile, working memory for calculation within each
0.94and the effect of intervention over place value column as well as for regrouping
control transitioned from significance to non- concepts and notation. We provided practice
significance at 2.79 SDs below the samples designed to build childrens accuracy and flu-
mean on language comprehension. Here, ency with these and the other strategies that
too, although effects were strong throughout we taught. Even so, children with stronger
the range of performance, findings working memory capacity demonstrated a
suggest that schools incorporate additional clear advantage. This suggests the need to
strategies to address the needs of specific incorporate additional instructional supports
childrenthis time, those with extremely to compensate for working memory deficits.
low language comprehension ability. Further- Moreover, the calculation intervention
more, researchers need to strengthen the exerted a strong main effect over the control
intervention to accommodate the needs of all condition (Powell etal., 2015). The present
children with MD. set of analyses extends this work by indicating
To identify directions for strengthening that the main effect favoring calculation over
the calculation intervention, it is important to control remains strong, even with a moderator
consider why childrens working memory effect and the main effects of seven other
and language comprehension might be asso- child-level abilities controlled in the model
ciated with responsiveness in this domain. and with the factor score incorporating proxi-
The calculation intervention focused on inter- mal and distal calculation outcomes.
connected knowledge about numbers, involv- The multilevel word-problem interven-
ing ideas and principles about whole numbers tion proved even more robust. For the word-
and the place value system. It also provided problem intervention, no moderator effects
children with counting strategies for addition were identified. The main effect favoring
and subtraction facts without regrouping. word-problem intervention over the control
Conveying these ideas and strategies group was strong throughout the distribution
depended heavily on the RA teachers and on all eight of the tested moderators, while
RA tutors verbal language. Although the the complex model also controlled the main
intervention design strived to encourage sim- effects of seven other child-level abilities or
ple, direct language even as it incorporated skills and when the factor score incorporated
the use of manipulatives and visuals, children word-problem outcomes.
with stronger language comprehension dem- These analyses provide the basis for con-
onstrated a clear advantage. This suggests the trasting child-level variables that predict
need to incorporate additional nonlanguage development in calculations versus word
instructional supports in strengthening this problemsacross environmental, instructional,
evidence-based intervention. It also suggests or intervention conditions. In investigating
that educators must incorporate adjustments these main effect predictors, our analyses are
for children with very low language compre- more stringent than most prior work because
hension. we controlled for the effects of a highly effica-
At the same time, executing the strategies cious intervention in our models, even as we
that we taught for deriving problem solutions restricted the range of performance on the
is demanding of working memory capacity. outcome by studying children with MD. Even
For example, in using the counting-up strat- so, attentive behavior emerged as a significant
egy to solve 9 minus 3, the child puts the 3 in predictor of both outcomes. Additionally,
his or her head (to remember to count up initial calculation skill uniquely explained
from 3) and then counts up to 9 holding up posttest calculation performance, while
one finger for each count (4, 5, 6, 7, 8, 9). language comprehension uniquely explained
The answer is the number of fingers used posttest word-problem skill.
374 Exceptional Children 83(4)

This further substantiates the importance careful training, support, and supervision to
of these abilities in mathematics develop- implement the interventions with fidelity. We
ment, as shown in prior work for attentive would recommend that schools provide ade-
behavior (e.g., L. S. Fuchs, Geary, etal., quate training and support to building staff as
2010), foundational calculation skill for later staff deliver these or any evidence-based
calculation performance (L. S. Fuchs etal., interventions. To the extent that this cannot be
2012), and language comprehension for ensured, overall impact and child responsive-
word problems (e.g., L. S. Fuchs, Geary, ness are likely to decrease. At the same time,
etal., 2010). In this study, every unit increase the interventions used in this study are fully
in attentive behavior was associated with a specified in manuals, with complete sets of
0.16-SD increase in later calculation skill materials available for implementation. This
and with a 0.33-SD increase in later word- provides the basis for wide distribution for
problem solving. Every unit in initial calcu- accurate implementation and for conducting
lation skill was associated with a 0.26-SD effectiveness studies, which are needed to
increase in the calculation outcome, and quantify intervention effects when school per-
every unit in language comprehension led to sonnel deliver intervention with the typical
a 0.20-SD increase in the word-problem out- levels of support available in the schools.
come. Second, although the estimates of ES from
the impact study were based on high-quality
randomized controlled trials, the moderator
Implications for Practice
analyses reported in the present analyses
As just noted, results from this study indi- were correlational. Therefore, conclusions
cated that teachers must be mindful of the about which cognitive processes moderate
possibility that children with extremely low the effects of the calculation intervention
working memory or language comprehension should be interpreted with caution. Third,
ability may require additional nonlanguage the targeted population was children with
instructional supports to optimize respon- concurrent difficulty across calculations and
siveness to calculation intervention. At the word problems. This population is important,
same time, it is important to note that atten- given that these learners experience inferior
tive behavior emerged as a significant overall outcomes than do children with difficulty in
(main effect) predictor even in the context of just one domain and that the majority of
highly effective calculation or word-problem children with MD experience difficulty
interventions that included evidence-based across these domains. Even so, it would be
self-regulation strategies. This suggests the potentially instructive to study the child-level
need to extend the typical kinds of strategies variables associated with responsiveness for
incorporated into evidence-based interven- children with initial difficulty in calculations
tions, perhaps with regular feedback to chil- versus initial difficulty with word problems.
dren on their learning progress and by Such a design may expand insights in impor-
motivating at-risk learners to use that feed- tant ways, and future research taking this
back to take additional responsibility for approach is needed.
directing components of their school pro- These study limitations notwithstanding,
gram, perhaps in terms of homework. we draw three major conclusions. First, in
both interventions (calculation and word
problem), effects were strong throughout the
Study Limitations and Conclusions range of initial pretest scores among children
The impact study (L. S. Fuchs, Powell, etal., with MD who experienced concurrent diffi-
2014; Powell etal., 2015) on which the pres- culty with calculations and word problems.
ent analyses were conducted was limited in Initial domain-specific performance did
three important ways. First, research staff, not moderate intervention effects for either
who were not certified teachers, delivered intervention. That is, neither initial calcula-
the intervention. Research staff did receive tion skill nor initial understanding about
Powell et al. 375

number was associated with childrens and Individual Differences, 42, 815824.
responsiveness to calculation intervention. doi:10.1016/j.paid.2006.09.018
Neither initial word-problem skill nor initial Bauer, D. J., & Curran, P. J. (2005). Probing
understanding about number was associated interactions in fixed and multilevel regres-
sion: Inferential and graphical techniques.
with responsiveness to word-problem inter-
Multivariate Behavioral Research, 40,
vention. This suggests strong efficacy for
373400. doi:10.1207/s15327906mbr4003_5
these interventions and indicates their added Branum-Martin, L., Fletcher, J. M., & Stuebing, K. K.
value within a multitier system. (2012). Classification and identification
Our second conclusion is that whether of reading and math disabilities: The special
domain-general cognitive abilities are associ- case of comorbidity. Journal of Learning
ated with responsiveness to mathematics Disabilities, 46, 490499. doi:10.1177/00222
intervention depends on the interventions 19412468767
focus: calculations versus word problems. For Cirino, P. T., Fuchs, L. S., Elias, J. T., Powell, S. R., &
children receiving a multitier calculation Schumacher, R. F. (2015). Cognitive and math-
intervention, working memory and language ematical profiles for different forms of learning
difficulties. Journal of Learning Disabilities,
comprehension were associated with respon-
48, 156175. doi:10.1177/0022219413494239
siveness to intervention. By contrast, we iden-
Connolly, A. J. (1998). KeyMath-Revised. Circle
tified no domain-general abilities that were Pines, MN: American Guidance.
active in influencing responsiveness to the Cook, B. G., & Cook, S. C. (2013). Unraveling
word-problem intervention. Future research evidence-based practices in special education.
may therefore productively focus on specific The Journal of Special Education, 47, 7182.
subdomain outcomes in the search for doi:10.1177/0022466911420877
moderators of responsiveness to intervention Cummins, D. D., Kintsch, W., Reusser, K., &
to provide insights on strategies for expanding Weimer, R. (1988). The role of understand-
the range of interventions available in the ing in solving word problems. Cognitive
schools. Psychology, 20, 405438. doi:10.1016/0010-
0285(88)90011-4
Every Student Succeeds Act of 2015, Pub. L. No.
For children receiving a multitier 114-95.
calculation intervention, working Fuchs, D., Compton, D. L., Fuchs, L. S., Bryant,
memory and language comprehension J., & Davis, G. N. (2008). Making sec-
were associated with responsiveness ondary intervention work in a three-tier
to intervention. responsiveness-to-intervention mode: Findings
from the first-grade longitudinal reading study
Finally, we reiterate that attentive behav- of the National Research Center of Learning
ior was the only domain-specific or domain- Disabilities. Reading and Writing, 21,
general variable that predicted performance 413436. doi:10.1007/s11145-007-9083-9
Fuchs, L. S., Compton, D. L., Fuchs, D., Powell,
across the intervention and control groups on
S. R., Schumacher, R. F., Hamlett, C. L., . . .
calculations as well as word problems. As
Vukovic, R. K. (2012). Contributions of domain-
already noted, this suggests the continued general cognitive resources and difference forms
importance of focusing on attentive behavior of arithmetic development to pre-algebraic
among children with MD and identifying knowledge. Developmental Psychology, 48,
methods to enhance self-regulation in this 13151326. doi:10.1037/a0027475
population. Fuchs, L. S., Fuchs, D., Compton, D. L., Powell,
S. R., Seethaler, P. M., Capizzi, A. M., . . .
Fletcher, J. M. (2006). The cognitive corre-
References lates of third-grade skill in arithmetic, algo-
American Psychiatric Association. (1994). rithmic computation, and arithmetic word
Diagnostic and statistical manual of mental problems. Journal of Educational Psychology,
disorders (4th ed.). Washington, DC: Author. 98, 2943. doi:10.1037/0022-0663.98.1.29
Barrett, P. (2007). Structural equation mod- Fuchs, L. S., Fuchs, D., Craddock, C., Hollenbeck,
eling: Adjudging model fit. Personality K. N., Hamlett, C. L., & Schatschneider, C.
376 Exceptional Children 83(4)

(2008). Effects of small-group tutoring with Fuchs, L. S., Zumeta, R. O., Schumacher, R. F.,
and without validated classroom instruction Powell, S. R., Seethaler, P. M., Hamlett, C. L.,
on at-risk students math problem solving: & Fuchs, D. (2010). The effects of schema-
Are two tiers of prevention better than one? broadening instruction on second graders
Journal of Educational Psychology, 100, word-problem performance and their ability
491509. doi:10.1037/0022-0663.100.3.491 to represent word problems with algebraic
Fuchs, L. S., Fuchs, D., Stuebing, K., Fletcher, J. equations: A randomized control study. The
M., Hamlett, C. L., & Lambert, W. E. (2008). Elementary School Journal, 110, 440463.
Problem-solving and computation skills: Geary, D. C. (2004). Mathematics and learning dis-
Are they shared or distinct aspects of math- abilities. Journal of Learning Disabilities, 37,
ematical cognition? Journal of Educational 415. doi:10.1177/00222194040370010201
Psychology, 100, 3047. doi:10.1037/0022- Geary, D. C. (2011). Cognitive predictors of
0663.100.1.30 achievement growth in mathematics: A 5-year
Fuchs, L. S., Geary, D. C., Compton, D. L., Fuchs, longitudinal study. Developmental Psychology,
D., Hamlett, C. L., & Bryant, J. V. (2010). 47, 15391552. doi:10.1037/a0025510
The contributions of numerosity and domain- Geary, D. C., Hoard, M. K., Byrd-Craven, J.,
general abilities to school readiness. Child Nugent, L., & Numtee, C. (2007). Cognitive
Development, 81, 15201533. doi:10.1111/ mechanisms underlying achievement deficits
j.1467-8624.2010.01489.x in children with mathematical learning dis-
Fuchs, L. S., Geary, D. C., Compton, D. L., Fuchs, ability. Child Development, 78, 13431359.
D., Schatschneider, C., Hamlett, C. L., . . . doi:10.1111/j.1467-8624.2007.01069.x
Changas, P. (2013). Effects of first-grade Geary, D. C., & Widaman, K. F. (1992). Numerical
number knowledge tutoring with contrast- cognition: On the convergence of componen-
ing forms of practice. Journal of Educational tial and psychometric models. Intelligence, 16,
Psychology, 105, 5877. doi:10.1037/ 4780. doi:10.1016/0160-2896(92)90025-M
a0030127 Groen, G., & Resnick, L. (1977). Can pre-school
Fuchs, L. S., Geary, D. C., Fuchs, D., Compton, D. children invent addition algorithms? Journal
L., & Hamlett, C. L. (2014). Sources of indi- of Educational Psychology, 69, 645652.
vidual differences in emerging competence doi:10.1037//0022-0663.69.6.645
with numeration understanding versus multi- Hayes, A. F. (2013). Introduction to mediation,
digit calculation skill. Journal of Educational moderation, and conditional process analysis:
Psychology, 106, 482498. A regression-based approach. New York, NY:
Fuchs, L. S., Hamlett, C. L., & Powell, S. R. Guilford.
(2003). Second-Grade Calculations Battery. Hayes, A. F., & Matthes, J. (2009). Computational
Available from L. S. Fuchs, 228 Peabody, procedures for probing interactions in OLS
Vanderbilt University, Nashville, TN 37203. and logistic regression: SPSS and SAS imple-
Fuchs, L. S., Malone, A., Schumacher, R. F., mentations. Behavior Research Methods, 41,
Namkung, J. M., Hamlett, C. L., Jordan, 924936. doi:10.3758?BRM.41.3.924
N. C., . . . Changas, P. (2016). Supported Hoover, H. D., Hieronymous, A. N., Dunbar, S.
self-explaining during fraction intervention. B., & Frisbie, D. A. (1993). Iowa Test of Basic
Journal of Educational Psychology, 108, Skills, Form K. Itasca, IL: Riverside.
493508. doi:10.1037/edu0000073 Jordan, N. C., & Hanich, L. B. (2000). Mathematical
Fuchs, L. S., Powell, S. R., Cirino, P. T., thinking in second-grade children with dif-
Schumacher, R. F., Marrin, S., Hamlett, C. L., ferent forms of LD. Journal of Learning
. . . Changas, P. C. (2014). Does calculation or Disabilities, 33, 567578.
word-problem instruction provide a stronger Kintsch, W., & Greeno, J. G. (1985). Under-
route to pre-algebraic knowledge? Journal of standing and solving word arithmetic prob-
Educational Psychology, 106, 9901006. lems. Psychological Review, 92, 109129.
Fuchs, L. S., Sterba, S. K., Fuchs, D., & Malone, doi:10.1037/0033-295X.92.1.109
A. (2016). Does evidence-based fractions Kline, R. B. (2016). Principles and practice of
intervention address the needs of very low- structural equation modeling (4th ed.). New
performing students? Journal of Research on York, NY: Guilford Press.
Educational Effectiveness, 9, 662677. doi:10. Lembke, E. S., Garman, C., Deno, S. L., &
1080/19345747.2015.1123336 Stecker, P. M. (2010). One elementary
Powell et al. 377

schools implementation of response to Vukovic, R. K., Fuchs, L. S., Geary, D. C., Jordan,
intervention (RTI). Reading and Writing N. C., Gersten, R., & Siegler, R. S. (2014).
Quarterly, 26, 361373. doi:10.1080/10573 Sources of individual differences in childrens
569.2010.500266 understanding of fractions. Child Development,
National Council of Teachers of Mathematics. (2014). 85, 14611476. doi:10.1111/cdev.12218
Principles to action. Reston, VA: Author. Wechsler, D. (1999). Wechsler Abbreviated Scale
Pickering, S., & Gathercole, S. (2001). Working of Intelligence (WASI). San Antonio, TX:
Memory Test Battery for Children. London, Pearson Education.
England: Psychological Corporation. Willcutt, E. G., Petrill, S. A., Wu, S., Boada, R.,
Powell, S. R., Fuchs, L. S., Cirino, P. T., Fuchs, D., Defries, J. C., Olson, R. K., & Pennington,
Compton, D. L., & Changas, P. E. (2015). Effects B. F. (2013). Comorbidity between reading
of a multi-level support system on calcula- disability and math disability: Concurrent
tion, word-problem, and pre-algebraic learning psychopathology, functional impairment,
among at-risk learners. Exceptional Children, and neuropsychological functioning. Journal
81, 443470. doi:10.1177/0014402914563702 of Learning Disabilities, 46, 500516.
Preacher, K. J., & Hayes, A. F. (2008). Asymptotic doi:10.1177/0022219413477476
and resampling strategies for assessing and Wise, J. C., Pae, H. K., Wolfe, C. B., Sevcik, R.,
comparing indirect effects in multiple media- Morris, R. D., Lovett, M., & Wolf, M. (2008).
tor models. Behavior Research Methods, 40, Phonological awareness and rapid naming
879891. doi:10.3758/brm.40.3.879 skills of children with reading disabilities and
Raghubar, K. P., Barnes, M. A., & Hecht, S. A. children with disabilities who are at risk for
(2010). Working memory and mathematics: mathematics difficulties. Learning Disabilities
A review of developmental, individual dif- Research and Practice, 23, 125136. doi:10.1111/
ference, and cognitive approaches. Learning j.1540-5826.2008.00270.x
and Individual Differences, 20, 110122. Woodcock, R. W. (1997). Woodcock Diagnostic
doi:10.1016/j.lindif.2009.10.005 Reading Battery. Itasca, IL: Riverside.
Riley, M. S., & Greeno, J. G. (1988). Developmental Woodcock, R. W., McGrew, K. S., & Mather, N.
analysis of understanding language about (2001). Woodcock-Johnson III tests of cogni-
quantities and of solving problems. Cognition tive ability. Itasca, IL: Riverside.
and Instruction, 5, 49101. doi:10.1207/ Zheng, X., Swanson, H. L., & Marcoulides, G. A.
s1532690xci0501_2 (2011). Working memory components as pre-
Siegler, R. S., & Booth, J. L. (2004). Development dictors of childrens mathematical word prob-
of numerical estimation in young children. lem solving. Journal of Experimental Child
Child Development, 75, 428444. doi:10.1111/ Psychology, 110, 481498. doi:10.1016/j.
j.1467-8624.2004.00684.x jecp.2011.06.001
Siegler, R. S., & Opfer, J. E. (2003). The devel-
opment of numerical estimation: Evidence for
Authors Note
multiple representations of numerical quantity.
Psychological Science, 14, 237243. This research was supported by Awards
Swanson, H. L. (2011). Working memory, atten- R01HD053714 and R24HD075443 and Core Grant
tion, and mathematical problem solving: A HD15052 from the Eunice Kennedy Shriver National
longitudinal study of elementary school chil- Institute of Child Health and Human Development to
dren. Journal of Educational Psychology, 103, Vanderbilt University. The content is solely the
821837. doi:10.1037/a0025114 responsibility of the authors and does not necessarily
Swanson, J., Schuck, S., Mann, M., Carlson, C., represent the official views of the Eunice Kennedy
Hartman, M., Sergeant, J., . . . McCleary, R. Shriver National Institute of Child Health and Human
(n.d.). Categorical and dimensional definitions Development or the National Institutes of Health.
and evaluations of symptoms of ADHD: The
SNAP and SWAN Ratings Scales. Retrieved Manuscript received March 2016; accepted
from http://www.adhd.net January 2017.

También podría gustarte