Mapping cognitive pathways in mastering long division: A case study of grade 5-6 learners

supported with a dynamic model of proximal assessment and learner diagnosis

Madhabi Chatterji and Nancy Koh

Teachers College, Columbia University

Howard Everson
Fordham University

Pearl Solomon
St. Thomas Aquinas College

For reprints, please contact:

Madhabi Chatterji, Ph.D
Associate Professor of Measurement and Evaluation
Co-Director, Assessment and Evaluation Research Initiative (AERI)
Teachers College Columbia University
525, W 120th St.
New York, NY 10040.
E-mail: [email protected]

Paper presented at the annual meeting of the American Educational Research Association on
March 28, 2008 at New York, USA

2
Mapping cognitive pathways in mastering long division: A case study of grade 5-6 learners
supported with a dynamic model of proximal assessment and learner diagnosis

Abstract
This paper presents a single-classroom case study of learners as they transitioned from grade 5-6,
documenting their learning gaps, changes in cognitive development, and mastery of one
mathematics domain, when supported with a Proximal Assessment for Learner Diagnosis
(PALD) model by their teachers. PALD is a dynamic approach to classroom assessment that
draws on the combined literatures in the cognitive and assessment sciences. In this study, we
focus on learners at three developmental levels transacting with situated problems in the Long
Division (LD) domain, a formal part of the mathematics curriculum at the school district
research site. The paper first documents the operational components of the PALD model,
showing examples of PALD practices used by teachers who taught the class in grades 5-6, and
provides the foundational theory. It then combines quantitative analyses and qualitative expert
reviews of students work to identify different ways in which children at different developmental
levels "think" as they gain expertise in the LD domain. While the class as a whole made visible
gains in one year, those who were developmentally most delayed initially were left behind. The
case analysis shows specifically where learning stalled or showed spurts, and how the PALD
approach can effectively support student learning, highlighting implications for math pedagogy,
diagnostic assessment, and teachers assessment capacities.

3
Mapping cognitive pathways in mastering long division: A case study of grade 5-6 learners
supported with a dynamic model of proximal assessment and learner diagnosis

Educational assessments that provide a close-up view of how children learn and develop
in a domain, incorporating up-to-date knowledge in the cognitive and learning sciences, are still
rare or absent in both classroom and large scale applications in the U.S today. A relatively recent
national academy report called for more research in this particular area, emphasizing that design
efforts directly engage teachers (Pellegrino, Chudowski, & Glaser, 2001). Coupled with that call,
for some time now, there has also been a cry for more student-centered and formatively-oriented
classroom assessment (Shepard, 2000; Stiggins, 2002; Wiggins, 1998).
The notion of formative classroom assessment, linked to instruction, and oriented
towards developing learner proficiency in defined areas, is not a new concept (see Bloom, 1982;
Gagne, 1997; Nitko, 1989). Over time, the spirit of formative assessment been evidenced in both
general and special education environments under various protocols and labels (see for
examples, Fuchs & Fuchs, 2003; Guskey, 2003). The idea has been revived today as a countermovement to high stakes testing accompanying the national push for educational excellence and
equitable outcomes under the NCLB Act of 2001. In recent applications, however, formative
uses of assessment by classroom teachers have relied too often on data from state-administered,
standards-based tests (on which schools are held accountable). Alternatively, such practices have
depended on locally-developed tests that mimic state instruments in terms of item content and
structure.
Externally-designed, standards-based tests rarely yield the fine-grained, diagnostic
information necessary for teachers to effectively detect learner gaps along developmental

4
continua as students grapple with and gain expertise in particular domains. Although higher
student performance might be evidenced on results of external tests as a consequence of the
popular practice of teaching to the test, such spikes in the data are temporary, and student gains
are rarely sustained when cognitive gaps remain unaltered.
There is thus, a high need today for research that focuses on developing models of
classroom assessment that are learning-centered, and that can concurrently develop in teachers,
the necessary assessment skills and attitudes to assess learners proximally and diagnostically,
close to where the teaching and learning processes unfold in classrooms every day. The aim of
reported research study, which forms one part of a larger research and development effort
supported by the National Science Foundation, is to address these very needs.

Purpose
This paper reports a descriptive case study showing how one class of elementary students
responded to a dynamic approach to classroom assessment titled, Proximal Assessment for
Learner Diagnosis (PALD), and how their learning gaps were addressed over a two year period.
The PALD approach, elaborated next, draws on the current and combined knowledge base in the
cognitive, learning and assessment sciences. The case study follows the same group of learners
over a two-year period as they transitioned from grade 5-6, documenting their cognitive
development and thinking in one mathematics domain, when supported with a Proximal
Assessment for Learner Diagnosis (PALD) model by their teachers. Specifically, the study
focuses on learners at three developmental levels transacting with situated problems in the Long
Division (LD) domain, a formal part of the mathematics curriculum at the school district
research site.

5
The larger research project on the PALD approach began 3 years ago (Chatterji &
Gordon, 2004), with field studies occurring between September 2005-June, 2007. Because
teachers hold the key to sound assessment implementation in classrooms, the PALD model was
formally field-tested and refined in partnership with an upstate New York school district, and
with the involvement of real teachers and students. Four elementary schools and 14 classroom
teachers were engaged as PALD participants in the 2-year long intervention development and
formative research project, along with 30 classrooms and teachers participating as Non-PALD
participants for comparison (Total Nstudents=700-800; Total Nteachers=44). To ensure that the
selected mathematics domains for the research project had local relevance, we drew on
mathematics content standards provided by the state of New York, as adopted by the district in
the form of curriculum maps. These maps served as guiding frameworks for designing projectrelated assessments for classroom and end-of-project use.
It is important to clarify at the start that the PALD research project incorporated a teacher
development component, focusing on changing teachers assessment attitudes and practices; that
component is not the focus of this paper. Rather, the present paper deals with learners cognitive
processes, learning gaps, and outcomes when exposed to PALD practices, using case study
methodology. Specific objectives of the present paper are as follows:
1) To describe the PALD model, present specific operational examples of how the PALD
model plays out in actual classrooms, and discuss the theoretical foundations of the same.
2) To document, using three different methods/perspectives, how thinking and expertise in
long division changed in one selected classroom of learners, as evidenced in their
responses to specially-designed, developmentally-diagnostic assessments administered at

6
the end of grade 5 and grade 6. (Developmentally-diagnostic assessments are specific to
the PALD approach).
3) To make inferences from the case data about learning and cognitive needs of learners at
three developmental levels as they transitioned through grades 5-6.
The case analysis shows how the PALD approach can effectively support student learning,
highlighting implications for math pedagogy, diagnostic assessment, and teachers assessment
capacities/skills.

Operational Elements of the Proximal Assessment for Learner Diagnosis (PALD) Model
PALD is characterized as a dynamic approach to classroom assessment. It calls on
teachers to make cyclic and frequent use of informal and formal diagnostic assessments, so as to
detect and address learning needs of students in selected domains, as they teach. According to
the model, teachers use of diagnostic assessment should be followed immediately with error
analysis, mediation (taking corrective actions tailored to learner needs) and provision of
intensive practice for learners to consolidate new learning and conceptual bonds.
Appendix A provides a diagrammatic representation of how the PALD model is expected
to be implemented by a teacher, with a theoretical sequence of activities in planning, teaching,
and concluding an instructional unit, marked as I-IV. The formative assessment and decisionmaking occurs during step III (PALD Implementation), and could involve as many cycles as
deemed necessary and feasible by the teacher. Step III can be repeated after new material is
introduced, with summative assessment (i.e., end-of-unit testing for assigning marks, making
decisions on student progress and promotion to next level or unit) withheld until students
learning gaps are sufficiently addressed. Individual teacher judgment is required with regard to

7
the duration, continuity/discontinuity, and number of PALD cycles that can be embedded within
a unit, based on the local curriculum scope and sequence, infrastructure, logistics, and schedules
at different schools.
In an ideal implementation, the PALD approach is dynamic, because it could potentially
involve adaptations and revisions to the plans for instruction, assessment, and to learning targets
set for a unit, to suit needs of learners progressing at different paces and in different ways as they
transact with a domain. This is indicated in the parts marked REVISE in the diagram shown.
The term domain has a specific connotation in the PALD approach. Instead of starting
with a listing of instructional goals and objectives (or content area standards found in typical
curriculum frameworks), the domain in the PALD approach emerges out of the backwards
analysis from situated, integrative and real-life tasks that teachers would like students to master
at the end of the lesson, unit, semester, or school year. Teachers connect a PALD domain with
broader curriculum frameworks as a part of the long-term planning process marked under step I
in Appendix A.
The focus of instruction and assessment planning in step II, centers around the type of
culminating tasks in which the teachers would like students to gain mastery and domain of
embedded concepts, skills, and competencies that flow out of the tasks. Box 1 below gives a
concrete example of a culminating task at end-of-grade 5 level, from which the Long Division
(LD) Domain was derived (focus of the present study).
Box 1. Culminating Task in Long Division Domain Used As Pre-Unit Assessment in a PALD Unit
Coconut cookies come in packets of 12. I would like to give 1 cookie to each 5th grader at (your school) to celebrate
the last day of school (in June). There are 119 5th graders at (your school). How many packets of cookies should I
buy so that every 5th grader gets at least 1 cookie? Will there be any cookies left over?
Show your plan to solve the problem. Then set up the numbers and solve it. Explain all the parts of your answer, and
show how you would check if your answer is correct. Show all your work.

8
Instead of being guided by abstract statements, such as, The student will apply a long
division algorithm/ procedure taught and other operations needed to solve problems with 4-digit
dividends and 2-digit divisors, PALD teachers would select concrete, culminating tasks likely to
be meaningful to learners as guides for designing of instruction and assessment. The
targeted/valued (math) concept knowledge, vocabulary, cognitive skills, complex procedures and
so on, are taught and tested through the medium of these situated real-life problems with which
students are likely to make a connection. De-contextualized teaching and testing of relevant
material (such as learning multiplication tables by themselves, outside a problem context), is not
altogether excluded; rather, it is kept to a minimum and treated as supplementary to the PALD
process (e.g., they can be used on the way to developing expertise in the culminating tasks that
are identified as the key learning targets for students).
The backwards analysis from targeted tasks during domain specification, is expected to
help teachers identify relevant types of prior knowledge that students need to attack the
culminating problems and to be successful. In applying the PALD process with teachers, we
asked them to envision and create such tasks, and to solve the problems first themselves so as to
be able to identify all embedded concepts/skills via a concrete method. The backwards domain
specification process is the means to help teachers identify the essential prior knowledge students
need, to attack a new domain. It leaves the door open for teachers to test, gauge and address
learners needs. By reinforcing, scaffolding, or teaching anew relevant but previously taught
concepts that learners need to grasp before moving to new material in the domain, they foster
learning development on a continuum (see PALD Implementation in step III in Appendix A).
As will become evident through concrete examples that follow in this section, the most
useful assessments for PALD practices, are not only domain-referenced, but they also have

9
properties that reveal where children get stuck as they struggle to learn something new or
different. They are labeled as developmentally-diagnostic assessments. Appendix B gives an
example of a multi-part, developmentally-diagnostic assessment in the LD domain.
Developmentally-diagnostic PALD assessments comprise tasks that are ordered by
complexity that tap backwards into students grasp of prior knowledge/skills relevant to gaining
mastery of a relatively novel domain of performance. Assessment formats used by classroom
teachers can be varied. They can be paper and pencil tests incorporating tasks such as those
shown in Box 1. They can involve use of probing Q & A that teachers embed as they teach and
build skills and concepts necessary for children to successfully solve the culminating
performance tasks that serve as guiding targets. They can incorporate structuredresponse
formats, or also include open-ended and performance assessments.
Regardless of format, to be diagnostically useful during teaching and learning, the task
design, prompts and accompanying scoring mechanisms must elicit or clearly reveal students
levels of understanding as well as misconceptions/ errors in a meaningful way to both teachers
and students. An example of a student- and teacher-friendly analytic rubric used to encourage
students to engage in self-monitoring their progress during a mini-unit on the LD domain follows
in Box 2. Analytic scoring provides a simple way for tracking error patterns.
As may be evident, in theory, PALD involves creating a safe and nurturing assessment
culture in the classroom, where learners are made aware of explicit learning targets. Affectively,
the learners must feel good as they engage in learning. In theory, thus, they should not be afraid
to closely examine their own learning gaps and errors, trusting their teachers to help them make
gains, fill gaps, and achieve greater levels of expertise in a domain. Within practical limits,

10
PALD is meant to be implemented in an ongoing manner by teachers, with students and teachers
jointly involved in detecting gaps and monitoring learner progress in targeted domains.

Box 2. A Student-friendly Rubric: Checking What I Know in Long Division (LD)

I can:
-Say in my own words what the LD problem is asking me to do
-Read a story problem and set up the LD and/or other operations needed to solve it
-Identify the dividend and divisor in an LD problem
-After the LD problem is set up, know how to start the algorithm (find digit with the highest place value in the dividend)
-Recall and use multiplication tables accurately when doing long multiplication and LD
-Recall place value concepts and give place value labels for digits in whole and decimal numbers, while doing long multiplication and LD
-Follow the steps of the LD algorithm correctly (for e.g., students use a mnemonic from class--DMSBAR)
-Repeat the LD algorithm until I get a remainder that cannot be further divided (grade 5)
OR
Continue the LD algorithm so I can express remainder as a decimal of the quotient (grade 6)
-Check the LD answer with backwards operations
-Explain what the answer means in my own words
-Think-aloud and write the steps of LD
-Think aloud and explain the answer and parts of the answer in a story problem
-Keep my scratch work neat (so that I dont get lost)
-Not be afraid to make mistakes
-Not give up before I am finished
-Teach LD to a friend
-Make connections between fractions, decimals and division/LD operations

Ideally, PALD data must also be generated proximally to learners, while instruction and
learning is in process--through probing and questioning during instruction, or appropriately
designed home-work, class-work, pre-unit or mid-unit exercises. A key part of the PALD process
is the timeliness and quality of follow-up actions taken by teachers, namely, mediation and
practice opportunities. Box 3 provides a sample of actual actions taken by a PALD teacher
during a 3-day unit on the LD domain with the class used for the case study.

11

Box 3. A Sample of PALD Activities during a Mini-unit on the LD Domain

1. On Day 1, the lesson began with a pre-assessment administered by the teacher. The reflections showed the following thread of
reasoning in the PALD lesson plan:
The first assessment I used was an integrated, real-life math task of mid-level difficulty on the domain continuum (coconut
cookie problem). It was the type of task on which we were shooting for mastery. My plan was to administer that task right at
the beginning and let the instruction flow from what we found there. The task was a 2-step, real-life word problem
involving several embedded competencies and concepts in long division, but no decimals. We were aiming for solving with
decimal numbers, but it was too early. Students would have to evaluate their answers after the long division procedure was
complete to see if the question in the problem was fully answered. If not, they would need a couple more operations to
complete the problem-solving. Interpreting different parts of the answer in the problem context was important for obtaining a
high score. The assessment was accompanied by some open-ended questions probing students reasoning, thinking and
feelings of confidence regarding long division.
2. Error analysis and review of the responses by the teacher(s) at the end of Day 1 showed an array of errors, but one specific area with
high levels of learner confusions/blocks.
[88% of the class of 17 students made] place value mistakes while doing the algorithmplacing part solutions in wrong places
and bringing down the wrong number from the dividend or getting stuck because of misplacements. Many were giving up after
this misstepleaving blank, scribbles or unfinished responses.
3. The teachers reflection and inferences about that subset of childrens cognitive and mediation needs were as follows:
[Need to] deepen understanding of what division is and how place value is related to bring down step of division; [they] need
better understanding of why we do long division of a 3-digit number in parts, starting from the digit with the highest place
value; why do we need to stay in columns when we bring down? What do digits in particular columns represent?
4. Mediation techniques that the teacher documented using were:
Use labels, H(undreds),T(ens),O(nes) ,1/10, 1/100, etc., as place value markers over columns of digits in the dividend (to
correct place value errors).
Think aloud the step in the algorithm and say why it was needed (to correct math reasoning errors), e.g., If 5 fits into the 7 in
700 one (1) time, we place the 1 or that part of the answer in the hundreds place of the quotient. Let us think--Why?
YesBecause it means we are multiplying 5 x 100 to get 500, sonow we can see whats left over when we subtract it from
700.. See?.. the 7 in 700 is also in the hundreds place.
Prompt and probe during recap/demonstration of problem-solving (to build math concept understanding and correct
vocabulary issues), e.g., When we do long division, we think of the divisor as a grouped number or a set. When you look at
a real life problem, we can look for word clues in the story to see if there is a grouped number or a set. In the cookie
problem, what is the grouped number that is most probably the divisor? Is it packets of cookies? So, what do we divide 119
by12?
5. Results from the follow-up practice assessment at the end of Day 1 (similar task with Orbit chewing gum packets, with 14 to a pack)
showed that the error rate on that concept/skill dropped to 50% in one day. PALD approaches continued through the end of the school
year. At the end, students were able to make nuanced statements about their own learning using the math vocabulary, such as:
[Problems] 14 & 17 they were the hardest because you had to divide numbers by 3 digit numbers and 2 digit numbers.

Theoretical Principles underlying the PALD Approach

A basic psychological premise on which PALD stands is the notion that cognitive and
affective capacities in relation to gaining expertise in given domains of performance, can be
modified through proximally-conducted, dynamic assessment and mediation. These principles
are adopted from a body of compelling work by Feuerstein and his colleagues in the late 1970s
who were able to show intellectual gains with this method in mentally retarded adolescents in
Israel (Feuerstein et al, 1979).

12
The ideas on cognitive modifiability are further reinforced by more current research in
the neurosciences suggesting that intellectual, affective and emotional capacities in humans are
not static, but can develop and change over ones lifetime (Goleman, 1994; 2007). Interventions
that forge and nurture mental connections or alter neural networks, can change behavior patterns
in humans, resulting in changes in our cognitive, emotional and social response patterns and
functioning.
Additionally, recent research on how individuals learn and gain expertise in particular
domains, synthesized by Pellegrino, Chudowski & Glaser (2001), reinforce Feuerstein et als
(1979) conclusions and broaden the base from which learning-centered assessment in classrooms
can be approached. The key elements are that (see Pellegrino, Chudowski & Glaser, 2001, p. 6163):

A persons prior knowledge, if made relevant to solving novel tasks/new kinds of

learning, helps construction of new knowledge and strengthens cognitive bonds

Scaffolded and facilitative instruction can help build new mental schemas, when learning
stalls or connections are not made by learners;

Problem-solving abilities and related learning are enhanced when learning tasks are
meaningful and situated, rather than de-contextualized;

Learning and problem-solving are integrative processes, where learners typically draw
on multiple cognitions--skills, concepts, and higher order abilitiesas relevant to a
given task;

Learners affective states and intellectual capacities are interdependent

13

Metacognitive capacities (i.e.,the ability of learners to think introspectively and

appraise their own understandings and grasp of a task/performance domain as they learn)
facilitates learning and development in given areas.

The proximal and dynamic characteristic of assessment for learning, surfaced as a necessary and
critical element in modifying learner outcomes in the above body of literature on cognitive and
developmental psychology.
The above principles were complemented with selected ideas from the literature on
domain-referenced testing, a close relative of criterion-referenced testing (CRT) methodology in
educational measurement, and from error analysis research in cognitive psychology (see Millman
& Greene, 1989; Nitko, 1989 for details on historical traditions of CRT; Nichols, 1994).
Documentation on how traditional CRT and domain-referenced testing has played out in current
state standards-based testing programs under NCLB, is also now available (see Linn, Baker, &
Betebenner, 2002). However, as already elaborated, PALD domains are different than traditional
connotations of test domains in domain-referenced test design.
PALD tests are designed from ordered domains, with the domain specification process
employing a backwards design procedure described. Additionally, the use of situated and
integrative tasks, and incorporation of probing and scaffolding within test design, are key
principles incorporated into the developmentally-diagnostic test design methods of the PALD
project that are not found in traditional CRT.
Further, traditional connotations of the term diagnosis, as used in standardized, skillbased diagnostic inventories are also different from PALD. Typical diagnostic tests, such as the
commercially-distributed Key Math Diagnostic Arithmetic Test, yield point-in-time indices of a
persons proficiency on different competency areas, showing profiles of strengths and

14
weaknesses of examinees by topical domains that report on status of an individual at a particular
time, such as Although the information is intended to have formative utility in instructional
contexts, in contrast to the dynamic PALD approach, these have been described in the literature
as static diagnostic tests.
In terms of the PALD instructional design, hierarchical or progressive lists of expected
outcomes and competencies with associated task analyses as the domain frameworks, are found
in the earlier work of Robert Gagne (1970). Such tests were often embedded in computerized
instructional modules, and tapped into pre-requisite skills and knowledge of students pertinent to
their attaining mastery of a domain. Feedback loops from such formative assessment were meant
to be closely tied to instruction, informing teachers on the effectiveness of their instructional
strategies as well as the adequacy and appropriateness of goals of instruction (Gagne, 1997).
Error analysis research offered ideas for building in the dynamic diagnosis and proximal
data use aspects of the PALD approach for teachers, as well as the test design elements that we
have labeled as developmentally-diagnostic assessments. Such work has been based on
qualitative and mathematical modeling of examinee error patterns has been conducted by both
measurement researchers and cognitive psychologists for some time now. Predominantly, the
research has relied heavily on large scale tests like the SAT or TIMSS. This fact led some
researchers to underscore the need for more as cognitively-diagnostic assessment design ideas
(see Nichols, 1994). Error analysis was never before directly connected with teachers and
pedagogy.

15
Method
We now move to describing in detail our case and case study methodology. In this
section, we provide our rationale for case and domain selection, characteristics of the teacher and
students in the case (along with two selected reference groups), the design specifications for
developmentally-diagnostic LD tests and empirically-determined properties of test items, and
discuss the rationale and procedures we used for the mixed-method analyses of the selected
classroom of students.
Case Selection
Of the 14 participating PALD classrooms in the research project, we chose one class that met
three criteria. These were:
(1) Students had been exposed to home-room teachers who were PALD-participants in both
grades 5 and 6 (or for two continuing years), and who were also their primary math
instructors at school;
(2) Teacher participants (and indirectly, the class) had received a majority of the PALD
intervention/coaching program and supports, including a 3-day demonstration unit on
PALD practices that focused on the LD domain; and
(3) There had been little or no attrition of students in the class over the two year period, with
the number of students with intact data from both Grade 5 and 6 LD assessments.
The class of 17 students met all these criteria. Only one student had been lost due to mobility,
leaving 16 cases for analysis. In this case, although the class had two different teachers in
grades 5 and 6, they were both PALD-trained and continuing participants in the project.
Selection of Mathematics Domains

16
Although the PALD intervention can be applied to any subject area, the present NSFfunded project focused on mathematics. We incorporated three units linked to local mathematics
curriculum frameworks (that were scheduled to be taught concentrically in both grade levels),
into the PALD intervention program. They were units on: long division (LD), geometry (angles
and triangles), prime factorization in the first year, and in the second year, algebra. When we
began the research, the leaders, staff and a local consultant informed us that LD is typically
taught to students at the district from grade 3 onwards. However, it is a topic with which students
continue to struggle even in grade 6. This information became a main motivation for focusing on
long division for in-depth testing of the PALD approach.
Student and Teacher Characteristics
Case. The teacher of the class in grade 6 was female and White. She held a Masters
degree in elementary education and had 9 years of teaching experience in all. She had the
reputation of being a good teacher, displayed an assertive, firm and confident classroom style,
with an outspoken manner outside. She was also the teacher union representative at the school.
The school had a good achievement history in the district on standardized tests.
The student group of 16 consisted of 3 (19%) Asian, 9 (56%) Black, 3 (19%) Hispanic, 1
(6%) White children, whose birth dates fell between January, 1995 through March, 1996. Of
these, 7 (44%) were girls. None were coded as English Language Learners (ELL) or Disabled on
the district database, but 9 (56%) were on the free/reduced lunch program (an indicator of
poverty).
During visits and the demonstration lessons, we found the class to be an active and
animated group. Students were not shy at all with strangers in the classroom. They soon became
aware of the purposes of our visits and, except for a couple who needed constant vigilance to

17
stay on task, fully participated in lesson-relevant interactions. Once in a while the class became
noisy, and their teacher needed to intervene to help manage classroom proceedings.
Reference Groups. To contextualize the results of the case in terms of the overall change
in LD domain scores, we also looked descriptively at two groups of Grade 6 students: those who
had had PALD-trained teachers for two continuing years in the four participating schools, and
those who had Non-PALD teachers for two continuing years in the same schools. It should be
noted here that the purpose for using the reference groups was not to examine comparatively
whether the class in the case study performed better or worse than the other groups, but to better
gauge what the typical change pattern was for same-aged children exposed to the same
curriculum, both with and without the PALD intervention. The characteristics of the reference
groups, described below, were very similar to the district demographic profiles for students and
teachers at the elementary level in terms of ethnicity and poverty.
The group of PALD teachers, including the teacher in the case, were 5 in number. They
were all White; all held Masters degrees; 2 were female; and their teaching experience ranged
from 4-11 years.
The total PALD student group at the end of grade 6 consisted of 8 (12%) Asian, 42 (62%)
Black, 13 (19%) Hispanic, 5 (7%) White children. The PALD students as a whole were older
than the children in the case, with dates of birth extending from December, 1993 through March,
1996. Of these, 33 (49%) were girls; 7 were coded as English Language Learners (ELL) and 3 as
Learning Disabled in the district database; 34 (50%) were on the free/reduced lunch program.
The Non-PALD teachers, including one substitute teacher, were 15 in number. They were
also all White and had Masters degrees; 12 were female. Other than the substitute teacher with
no experience, teaching experience here ranged from 4-23 years.

18
The Non-PALD student group consisted of 2 (1%) American Indians, 16 (7%) Asian, 143
(65%) Black, 38 (17%) Hispanic, and 22 (10%) White children. These students were also older
than the children in the case, with birthdays extending from May, 1993 through November, 1995.
Of these, 122 (55%) were girls; 27 were coded as English Language Learners (ELL) and 15 as
Learning Disabled and 4 with other disabilities on the district database; 84 (38%) were on the
free/reduced lunch program.
Data Source
The data source consists of student responses to structured-response and open-ended
items on two different but parallel forms of developmentally-diagnostic tests tied to the LD
domain, that were administered to the group of students at the end of grade 5 (Year 1, or with
one years exposure to a PALD-trained teacher) and end of grade 6 (Year 2, or with two years
exposure to PALD-trained teachers), respectively. Each of the 16 children in the case had a pair
of tests that showed a picture of their development over time. As indicated earlier, Appendix B
shows the details of the test design specifications, sample items, and estimated difficulty and
discrimination levels of individual items and integrated item sets, in Years 1-2.
Test Design Properties. Because the test design component is integral to the PALD
model and a necessary part of generating the type of data we think is necessary to support
learning and learner development, we conducted some validity examinations with the test data
using the class and the combined reference groups. The LD tests in both grades were parallel in
terms of math content and skills tapped, with the Grade 6 test extending the application of LD
procedure with decimal numbers. There were four components in each, labeled as Parts 1-4, that
are organized progressively in terms of intended complexity. The first item set is described in
detail to illustrate how the underlying test design principles were applied (see Appendix B).

19
Part 1 is an item set. It presents a numeric problem where students have to apply the LD
procedure. It starts with structured items probing students understanding of the LD operations,
and taps into some vocabulary knowledge situated in the problem presented (using math
vocabulary and language is emphasized in districts curriculum that was our research site.) The
probes lead students to the problem to be solved, presented in an open-ended format. The
questions are presented in a straight forward manner.
As during PALD implementation in the classroom, probing is embedded in the item set
(Part 1). The items are situated rather than de-contextualized, in that a problem is presented
first to which the probes directly refer (thus, math vocabulary and concept understanding are not
tested separately but tied to a problem that students have to ultimately solve). Both structured
and open-ended formats are used; the former is treated as a form of scaffolding, as the prompts
are intended to draw the examinees attention to specific aspects of the problem that they are
expected to attack thereafter.
If Parts 1-4 in Appendix B are examined together, the cognitive/learning science
principles incorporated in the test design are as follows: where possible, use of contextualized,
situated questioning; use of embedded probing; use of inter-connected item sets that become
progressively more complex; more difficult tasks calling for integrated use of multiple kinds of
concept knowledge, operations, rules/procedures, or problem-solving strategies; and the
possibility of both analytic (part-scoring) and holistic scoring on individual and related sets of
items. The items are intended to guide the students towards thinking about and attacking the
problem in a way consistent with the way they might typically be taught. The tests can be used in
a grade-free manner, following children as they develop in a domain, irrespective of their grade
level at school.

20
Psychometric Evidence. During construction, the tests and scoring rubrics were reviewed
for content validity by teachers, school district curriculum specialists and math education
consultant on the project, along with members of the research team. As the overall LD domain
score was used to make certain inferences about childrens change, and the expectation was that
all items would homogeneously tap into the same proficiency area, we also examined validity
and reliability with some traditional psychometric methods.
The total LD domain scores yielded internal consistency reliability estimates of .84
(grade 5) and .85 (grade 6), shown in Table 1. For the end-of-year scoring, raters were trained to
use the rubrics with practice test papers in workshops. The inter-rater agreement on individual
items/item sets in the first year ranged from .58 to .83, with a median value around .70. The
rubrics and procedures were further examined on the basis of the inter-rater study and workshop
discussions, and the scoring process improved considerably in Year 2 based on rater reports.
In terms of convergent validity with external measures, the LD total scores correlated
with scale scores of the New York state standardized math tests in the order of .62 in grade 6.
They correlated with internal PALD project measures in the order of .52 (geometry domain test
scores), .55 (algebra domain test scores), and .38 (scores from a self report survey tapping math
self-efficacy). These results together affirm the meaningfulness of the LD scores in terms of
tapping some overlapping math capacities and math-related affect. The convergent validity
estimates were obtained from a data set of 381 cases with complete data on all variables,
collected in Year 2 of the PALD project.
The above was necessary evidence for us to support use of the LD parallel tests. Of more
direct relevance to the PALD design, however, was the validation of the logic for ordering tasks
by complexity, using empirical difficulty and discrimination statistics on individual items and

21
ordered item sets. Would the estimated difficulty increase as intended, with items/item sets
designated as more complex? Also, on individual items/item sets, would the estimated difficulty
change between grade 5-6, following an added year of student exposure to math teaching and
PALD? The overall results for the class of 16 are shown in Appendix B. Item difficulty here
refers to percent of students who gave a correct response to an item. In the case of items with
rubrics yielding data on 2-7 point scales, we used the mid-point of the scale (rounded) to obtain
item difficulty estimates. The item discrimination indices are adjusted item to total score
correlations.
With a couple of exceptions (discussed in-depth in the Results section from a student
learning angle), the general ordering of tasks was affirmed. However, the rather high and
positive discrimination indices in the item statistics also raised a red flag, suggested that within
the class there were major differences in levels of expertise and performance on several
items/item sets, when high and low performers were sorted. This prompted a need for more indepth examination of childrens work.
Analysis of Student Responses: Cognitive Pathways of Learners
To begin the second and main phase of the analysis, children were sorted at three
different levels of performance at the end of grade 6, based on the distribution of total LD
domain score using total project samples testing results on the LD Test in 2007 (N=698). We
used the larger sample to derive reliable cut-scores. The three developmental levels (1-3) were
based on percentile ranks of 33 (raw score 27) and 66 (raw score 36) on the Year 2 test.
Applying these cut-scores to the class for this study, we had the following breakdown:
Level 1= 5 students (lowest performance level)
Level 2=4 students

22
Level 3=7 students (highest performance level).
We then went back to 2006 (end-of-grade 5) to retrieve pairs of tests for each child, reflecting
their status and change on the LD domain across two points in time.
We use three methods to analyze the class data. First, the quantitative analysis uses
descriptive statistics, including means, standard deviations and skewness measures to examine
distributions of the class and the two reference groups over time. Changes on the overall domain
score are documented using pooled standard deviation units. For the class as a whole, we also
examine the item analysis results against the concepts/abilities tapped by particular item
sets/items. Then, we take a look at the three developmental levels and study differences in
patterns of change on each part of the LD domain (Parts 1-4).
Second, we support the above qualitative error analysis of the same childrens work
based on expert reviews from cognitive psychology and math education perspectives. The error
analysis compares qualitative and quantitative summaries of errors made by children organized
at three developmental levels at the end of Year 2, without the scoring rubrics.
Third, we compare and triangulate the results of all three sets of analyses to map
cognitive pathways of learners at three developmental levels. Cognitive blocks and pedagogic
needs of learners are inferred.

Quantitative Case Study Results

Quantitative Analysis for Whole Class. Learning Change based on LD Domain Scores
This section will describe findings displayed in Tables 1-3 and Figure 1. We begin with
Tables 1-2, showing the overall performance of the class at the end of grade 5 and grade 6, on
the LD domain score. At the end of grade 5, the class distribution had a strong negative skew (-

23
.67) as may be expected with mastery-oriented approaches to teaching and assessment, like the
PALD. This meant that a majority were doing well, but a few were behind. The class mean
score was 23.72 (SD=10.67), and possibly influenced by low scorers. As shown in Table 2, the
class mean was placed .18 standard deviation units below the mean for the pooled sample in
grade 5.
The PALD reference groups distribution had no skew, and the mean was a little higher
in grade 5 at 24.12 (SD=11.47) than our case. The Non-PALD reference group in grade 5 had a
smaller negative skew at -0.437 than the case, and the mean was well above both the class and
the PALD group at 26.24 (SD=11.17).
In grade 6, the class distribution turned around in sharp contrast to both the reference
groups. The negative skew was higher in value at -.87, with a mean score of 32.12 (SD=14.66).
The PALD reference group in grade 6, which included the case, had a mean of 26. 02
(SD=10.52), in contrast. The PALD reference group had a clear negative skew of -.645 in the
second year, suggesting that there were higher numbers of students at the top now, with greater
measured levels of domain mastery. The Non-PALD had a slightly higher mean that the PALD
group of 26.47 (9.34), with the skewness measure again at -.435. Both reference group means in
grade 6 are lower than the class mean, when the case data are isolated.
In Table 2 we see that the class performance in pooled standard deviation units in grade
6 is +.59 SD units, relative to the pooled sample of 232 students. The class gain on the LD
domain registers at over +.40 SD units. The two reference groups do not look different from each
other, both located at the pooled group mean, but the larger PALD group as a whole registers a
small gain at +.11. The above analyses also showed that change statistics, rather than point-in-

24
time mean estimates, are better for gauging subtle and large shifts in overall domain mastery.
Comparative mean analyses mask the picture on learning development.
Quantitative Analysis for Whole Class. Learning Change based on Item Analyses
Looking at the results in Appendix B more closely, it becomes possible to look at how
the class performed based on concepts/cognitive processes tapped by particular items/item sets.
Because the class N is 16, item statistics may not reliable for evaluating test properties. However,
they are employed here to study where learning changed for the class and where it appeared to
stall. The very high positive discrimination indices at the item level, particularly, explain the
reason for the sharp negative skew in the grade 6 distribution for the class/case under study.
We find in Appendix B that the first LD concept understanding probe (#1) is relatively
easy at the end of grade 5 with a difficulty estimate at .75. It discriminates positively between
high and low performers at +.26. As expected following another year of intervention, this type of
item becomes much easier for the class as a whole in grade 6 with difficulty at .94. But, the type
of item now discriminates more between high and low performers with a discrimination value of
+.43.
In contrast, the open-ended item tapping the same kind of LD understanding (asking for a
restatement of the problem), is considerably harder for the class in grade 6 with difficulty at .75,
than a structured item tapping the same ability (#1). Although these were expected patterns, we
find that loss of scaffolding in an open-ended item, makes the task harder for many children in
the class, yielding a discrimination value of +.63.
Also in contrast, the item difficulty the LD vocabulary item (item type # 3, tapping recall
of concept knowledge) remains unchanged over time, and is relatively easy in both years at .87.88. As intended in the same item set, the application of the LD procedure (item type #6, Part 1,

25
numeric problem) in grade 5 is more difficult than the preceding items tapping concept
knowledge and understanding at .63. The same item becomes easier in grade 6 at .75. In both
years, we find that the application task is highly discriminating in terms of performance of
students who can and cannot do LD (D values of +.83 and +.85).
As intended, the most difficult item for the class was Part 4 (real-life problem requiring
use of decimals) at .50. This item was added to the developmental domain in grade 6. It also
discriminated positively and strongly at +.69.
There were two unexpected results in terms of how the tasks (test domain) interacted with
the class over time, shown in test results to Part 2 and Part 3 in Appendix B. On Part 2, we see
the items becoming slightly more difficult in grade 6 (drop of difficulty from .87 to .81). On Part
3, dealing with real-life problem-solving, a major target set for grade 5, we see no change in the
overall item difficulty or discrimination values (.75 difficulty, +.94 discrimination). This last
finding suggests that while the class achieved some level of mastery in grade 5 on this type of
task, there was no development beyond that point.
Quantitative Analysis by Developmental Level. Learning Change on Items/Item Sets
To identify learning pathways, Table 3 may now be examined by looking across the three
developmental groups performance by item/item set. The table identifies with bold and italics
when changes occur suggesting gains from grade 5-6, on each item set (Parts 1-4) of the LD
domain.
On the initial item probes in Part 1, we see some gains made by the children at the lowest
level. This group (Level 1) shows difficulty with the open-ended item asking for restatement of
the LD problem even in grade 6 (only 40% get it right), but they are able to handle similar probes

26
when scaffolded with a structured item format. The above conceptual difficulty disappears at
developmental Levels 2 and 3, irrespective of item format.
On the multi-step LD problem in Part 1, we find no change in the distributions in grade
5-6 at Level 1. However, we do see shifts at Levels 2-3 between the two grades, with all the
children becoming successful in grade 6 (2 children were unsuccessful at grade 5).
A similar pattern is seen in shifts on Part 2, dealing with checking of LD solutions. No
change is seen in the lowest level children. The children at higher developmental levels have
reached a level of mastery in grade 5; and this level is sustained in grade 6. The qualitative
reviews of responses next, suggest why the lowest level children remain stagnant between the
two grades, and the transitioning cognitive state relevant to LD mastery.
On Part 3, we see a small movement at Levels 2-3 to reach high levels of mastery in
grade 6. On this item set, one child at Level 1 shows a movement towards mastery.
Part 4, dealing with use of decimals during solving real-life LD problems, is clearly
found to discriminate between developmental Levels 2-3. The children at highest level have
reached mastery on this, although this material was introduced in grade 6; however, both lower
level groups are still unsuccessful.
Figure 1 displays the change in the skew in the shape of distributions by developmental
level on the overall LD domain, an indicator as to whether students were moving towards
mastery, and becoming more homogeneous. The distribution is positively skewed at Level 1 in
both years (most students at the bottom). Except for a small change in the middle bar moving
right, or towards mastery, there is no significant change. There are visible changes at Levels 2
and 3, where in grade 5 we see negative skews, leading to more homogeneous distributions at the
higher end of the scale in grade 6.

27

Qualitative Review of Selected Student Responses

Figures 2 and 3, representing samples of student responses at Level 1 and versus Level 3,
now provide an added picture of mental blocks of a child who is unsuccessful and one who is an
expert on the LD domain in Year 2. They suggest how learning might shift from one level to the
next, with appropriate mediation techniques.
We see in Figure 2 the paper of the only child who was partly successful on the real-life
problems at Developmental Level 1 in grade 6. She starts the LD algorithm, but cannot continue
it and reverts to drawing groups of circles to represent the divisor (clusters of 12 circles). She
manages to add enough clusters to get 167 (the dividend) and finds the remainder. This approach
to doing division is taught in schools in lower grades when it is first introduced. The student is
unable to progress beyond the concrete modeling approach to solving LD problems, despite
being taught the LD algorithmic procedure at school. We see she gets the right answers. She
interprets the answers in context. But she makes no connection at all with the LD algorithmic
process and her approach to division. Her interpretations are thus partially off-the-mark.
The paper in Figure 2 was chosen because it suggests a transitional stage in development
between learners who are completely stuck with LD at Level 1, and learners at Level 2 who
appear to have mastery in this type of problem. If we see Figure 3, in contrast, it is clear that this
child is at the highest developmental level on the LD domain.
The steps and details in working out the LD procedure are clearly displayed, scratch work
shown, and the verbal explanations are on point. Along with this we see the stronger ability of
the student to explain steps in writing---thinking aloud and communicating the process. We
thus become instantly aware that we are looking at the work of an LD expert in grade 6.

28
Both the quantitative and qualitative reviews of student responses were revealing, and fit
together in terms of telling a story about how the students developed over the period of our
study. However, we still needed a pedagogical perspective on learning development, keeping in
mind how mathematics, and LD in particular, is taught to learners at the intermediate level. The
next section provides these results. (Note to Readers: Detailed analyses in this section
forthcoming from our co-author, a cognitive psychologist).

Qualitative Analysis from a Math Pedagogy Perspective

Domain Description from a Teaching Perspective
The focus of the LD domain is on the application of the long-division algorithm to the
solution of real-life division problems. Although there are two categories of division problems,
partition or quotition (sometimes referred to as measurement division), the algorithm is usually
taught in terms of quotition. The reason for this is that in a quotition problem you are seeking
the number of groups of a given sized whole. The algorithm works more easily with the
commonly taught concept approach of, How many groups of this size fit into this number? or
at a deeper level of understanding, How many groups of three tens fit into nine tens? In a
quotition problem you are given the size of the group and solving for the number of groups. The
inverse (backwards) relationship of quotition to multiplication is also clear. How many groups
of size three are there in 15? is the inverse of the multiplication concept within, What will be
the size of the whole if there are five groups of 3?
In contrast, in a partition problem you are given the number of groups or parts and
seeking the size of each part or If I divide 15 into five parts how large will each part be?
Although the algorithm works for both forms and students quickly learn that the larger number

29
must be the dividend, the interpretation of the quotient and remainders for problem solving are a
greater challenge.
The algorithm has been traditionally taught by rote memorization of the sequence of steps
(sometimes with an acronym); divide, multiply, subtract, bring down and again or repeat. More
recently, greater understanding of the process has been added in the presentation of long division
as a sequence of steps in which the units of each place value from the greatest to the least are
divided and remainders regrouped to the next smaller place value. In either case, the ability to
identify and understand the meaning of dividend, divisor, quotient and remainder is a sometimesoverlooked but necessary basic to the ability to solve a given algorithm or construct the
algorithm from a word problem.
Qualitative Identification of Errors in Grade 5 Responses
Developmental Level 1. The only question answered correctly by most students was item
(1), which was structured-response that asked them to choose the meaning of an illustrated
division problem. The students also seemed to be able to identify the rule for division by five,
but there were several errors in the facts. Three students were able to identify the first step from
structured item. Although two students at Level 1 were able to set up the algorithm, all were
unable to complete the algorithm correctlyeven with a single digit divisor. The errors included:

knowledge of the facts (multiplication and division)

subtraction process

incorrect placement of partial products as well as the sequence of steps

Developmental Level 2. All students at Level 2 identified the divisor and dividend and
division by five rule. Three out of four of the students were able to identify the meaning of the

30
illustrated algorithm, where to begin, and correctly performed LD with a single and double digit
divisor, identified the remainder and constructed the steps taken.
Developmental Level 3. The only noticeable problem encountered by the Level 3 group
was in choosing that the first step was to look at the dividend from the number on the left. Only
2 students in the group of seven at this top level chose the right answer--yet they all performed
the algorithm correctly. Two of the group did not precisely describe the steps they performed for
the algorithm. One omitted the item and the other merely abbreviated the numerical calculations
Qualitative Identification of Errors in Grade 6 Responses
Developmental Level 1. There was some evidence of the ability to identify the divisor and
dividend. Most were also able to identify the divisibility rule and choose a first step, which was
expressed as a place value. The entire group, however, could not correctly construct the sequence
of steps even though the names were listed. Only one student of the group was able to complete
the algorithm correctly. None could construct and complete the algorithm from the problem.
Developmental Level 2. All students were able to identify divisor and dividend, division
by 2 rule (most with an example), correctly perform the single digit, no remainder algorithm and
with one exception list the steps in order. Interestingly, most of the students in this group were
also able to accurately construct a two-digit algorithm from the problem given and correctly
complete it, but they had some difficulty in identify the quotient and in constructing (in words)
the meanings of divisor, dividend, and remainder. Only one student correctly checked the
quotient with the inverse multiplication. The most difficulty for the group was in setting up the
partition problem and converting the remainder to a decimal.
Developmental Level 3. The question on this test about where to begin LD was phrased in
terms of highest place value, but two of this top group still made the wrong choice. All set up the

31
algorithms correctly and completed them with single and double digit divisors and decimal
conversions of remainders. An interesting observation for this group is that when they had a LD
problem with remainders they automatically converted the remainders to decimals and
sometimes rounded the quotient to the next whole number. This sometimes confounded the
expected explanation of the problem solution.
Inferences about Pedagogic Needs
Student responses show that the ability to correctly perform the rote algorithm sometimes
seems to be separated from its connection to a problem to be solved. When performed without
understanding, LD is less of a challenge. This is the pattern at Level 2. The items in the
assessments for each year represented a sequence of content difficulty levels from a pedagogy
perspective, with ranges including: single digit to double digit divisors, quotients with and
without whole number remainders to conversion of remainders to decimals, and un-constructed
to open-ended constructed responses.
For the lowest developmental level, there is a need for recursive review of the division
facts, meaning of division, and steps in the algorithm as well as practice in applying the
connected vocabulary and using it to describe the process. They still definitely need some
individualized support using manipulatives and direct connections between these, the problems,
and the algorithm.
Students at Level 2 may need more practice with setting up partition vs. quotition
problems as well as decimal applications. The processes within the algorithm as well as the
interpretation of the quotient as a solution to the problem are critical. Application of the inverse
multiplication as a check, is also needed practice.

32
It is obvious that the Level 3 group has mastered the algorithm process in grade 6.
Recursive review of further applications to both quotition and partition real-life problems may be
helpful in firming up long-term memory and preparation for applications to algebraic
expressions.

Conclusions
The case in this paper showed how children attained expertise in the LD domain, using
results from a particular type of diagnostic testing and test design integral to the PALD approach,
coupled with methods in mathematics pedagogy prevalent at the district research site. Results
identify different ways in which children at different developmental levels "think" as they
improve in long division, suggesting how the PALD approach can be used to further support
their learning.
The class made striking gains between grade 5 and 6 on the LD domain. However, those
at the lowest developmental level were left behind. Domain-specific gaps became very clear
through a combination of descriptive quantitative analysis and qualitative reviews of responses;
further types of mediation and scaffolding that might be effective are also suggested in the
results.
In terms of cognitive pathways, we find that students at the lowest Level 1 had not
mastered the LD algorithm by and large over the two year period, but were able to respond when
scaffolded, to probes on LD understanding. There appeared to be a transitional stage whereby
they perform division with understanding, but using concrete pictorials/manipulatives. To make
the next cognitive connections to use of the LD algorithmic process, they need further scaffolded
instruction and assessment.

33
The mid-level students at Level 2 appear to master the algorithmic process, but they
might often do it by rote (without understanding or the ability to explain). The learning and
mediation needs for this level of learning are more practice with reasoning, thinking aloud and
explaining. The LD experts at Level 3 gain the ability to articulate and communicate what they
think, along with masterful and clean application of the procedure.
We found a good deal of agreements in the quantitative analysis with data from the
developmentally-diagnostic tests and the qualitative observations and reviews of student growth.
This triangulation may be viewed as a validation of the evidence generated from the tests.
The types of quantitative and qualitative analyses illustrated are not overly complex, and
may be meaningful to most teachers. Thus, a next step might be in incorporating materials from
case studies such as this one in teachers professional development materials. Anecdotally, we
can report that the teacher in our case study communicated serious reservations about the time
and work the PALD approach entailed in the first year of the intervention research. At the end of
the second year, she specifically asked to continue using the PALD approach in the following
academic year, despite the fact that the grant-funded project had reached closure and there would
be no funding/supports available.
This study is significant in that it adds to the theoretical knowledge base on cognitive and
assessment sciences as it impacts classroom teaching practices and learner development,
successfully marrying these areas, as recommended in recent national academy reports
(Pellegrino, Chudowski & Glaser, 2001). The ideas for developmentally-diagnostic test design as
presented here, marry the highlighted ideas from psychological literature, with selected
principles from the assessment sciences. Further research and development on the PALD model
is continuing.

34
References

Bloom, B. S. (1982). Human characteristics and school learning. New York, NY: McGraw-Hill
Book Co.
Chatterji M. & Gordon, E.W. (2004). Improving mathematics achievement in grade 5-6 students
with systemic use of proximal assessment data. New York, NY. Teachers College
Columbia University. Proposal supported by the National Science Foundation, EREC
solicitation. December, 2004.
Feuerstein, R. et al (1979). The dynamic assessment of retarded performers: The learning
assessment potential device, theory, instruments and techniques. Baltimore, MD:
University Park Press.
Fuchs, L.S., & Fuchs, D. (2003). Can diagnostic reading assessments enhance general educators
instructional differentiation and student learning? In B. Foorman (Ed.) Preventing and
remediating reading difficulties: Bringing science to scale (pp. 325-351). Baltimore,
MD: York Press.
Gagne, R. (1970). The conditions of learning (2nd. Ed.). New York, NY: Holt, Rienhart &
Winston.
Gagne, R. (1997). Mastery learning and instructional design. Performance Improvement
Quarterly, 10(1), 8-19.
Goleman, D. (1994). Emotional intelligence: Why it can matter more than IQ. New York, NY:
Bantam.
Goleman, D. (2007). Social intelligence: The new science of human relationships. New York,
NY: Bantam-Dell.

35

Guskey, T. R. (2003). How classroom assessments improve learning. Educational Leadership,

60(5), 6-11.
Linn, R.L., Baker, E. L., & Betebenner, D. W. (2002). Accountability systems: Implications of
requirements of NCLB Act of 2001. Educational Researcher, 31 (6), 3-16.
Millman, J., & Greene, J. (1989). The specification and development of tests of achievement
and ability. In Educational Measurement by R. L. Linn (Ed.), pp. 335-366: American
Council on Education.
Nichols, P. D. (1994). A framework for developing cognitively-diagnostic assessments.
Review of Educational Research, 64 (4), 575-603.
Nitko, A. J. (1989). Designing tests that are integrated with instruction. In R.L. Linn (Ed.),
Educational Measurement (3rd ed., pp. 447-474). New York: American Council on
Education and MacMillan Publishing Co.
No Child Left Behind Act of 2001 (NCLB), Public Law No. 107-110, 115 Statute 1425 (2002).
Pellegrino, J.W., Chudowski, N., & Glaser, R. (Eds). (2001). Knowing what students know: The
science and design of educational assessment. Washington, D.C: National Academies
Press.
Shepard, L.S. (2000). The role of assessment in a learning culture. Educational Researcher,
29(7), 4-14.
Stiggins, R. J. (2002). Assessment crisis: The absence of assessment for learning. Phi Delta
Kappan, 83(10), 758-765.
Wiggins, G. (1998). Educative assessment. San Francisco: Jossey-Bass.

36

Table 1. Performance of Class Selected for Case Study versus Student Cohorts* on Long Division (LD) Domain:
Descriptive Statistics in Year 1 and Year 2 on LD Total Scores
Year/Grade

Group

Mean

SD

Max
Score

Min
Score

Skewness

Kurtosis

Standardized
Alpha
Reliability
for LD Score

Item-level
Inter-rater
Agreement:
Range
(median)

2006/ End of
Grade 5

Class

23.72

10.67

34

-0.672

-1.278

16

.84

.58-.75 (.70)

PALD
Cohort

24.12

11.47

43

-0.058

-0.981

55

NonPALD
Cohort

26.24

11.17

43

-0.437

-0.818

161

Class

32.12

14.66

47

-0.866

-0.721

16

.85

Pending

PALD
Cohort

26.02

10.52

40

-0.645

-0.378

55

NonPALD
Cohort

26.47

9.34

40

-0.435

-0.553

161

2007/ End of
Grade 6

* Note: PALD student cohort (includes class for case study) was taught mathematics for two continuing years by
PALD-participant teachers. Non-PALD cohort was taught mathematics for two continuing years by non-participant
teachers. Raw scores not comparable in Year 1 and 2 due to different test lengths.

37
Table 2. Grade 5-6 Gains in Pooled Standard Deviation Units on Long Division (LD) Total Scores: Class Selected
for Case Study versus Student Cohorts*
Year/Grade

Group

2006/ End of
Grade 5

Class

-.18

16

PALD Cohort

-.15

55

Non-PALD
Cohort

+.04

161

2007/ End of
Grade 6

Mean

SD

Mean z-score

Mean
z-score GainYear 2

+.59

+.41

232
16

PALD Cohort

-.04

+.11

55

Non-PALD
Cohort

+.00

+.00

161

Pooled
Class

Pooled

25.79

26.44

11.12

9.67

232

*Note: PALD student cohort (includes class for case study) was taught mathematics for two continuing years by
PALD-participant teachers. Non-PALD cohort was taught mathematics for two continuing years by non-participant
teachers.

38
Table 3. Item-Level Performance of Sub-Groups by Developmental Level on Long Division (LD) Domain:
Changes* from Year 1- Year 2
Task Set

Task Descriptor

Score
Levels

Developmental Level 1

Developmental Level 2

Developmental Level 3

Grade 5

Grade 6

Grade 5

Grade 6

Grade 5

Grade 6

f (%)
2 (40%)

f (%)
1 (20%)

f (%)
1 (20%)

f (%)
-

f (%)
1 (14%)

f (%)

3 (60%)

4 (80%)

3 (75%)

4 (100%)

6 (86%)

7 (100%)

Open-ended item (0/1)

2 (40%)

Situated use of LD Vocabulary

(1)

2 (40%)

2 (40%)

3 (60%)

3 (60%)

2 (40%)

1 (20%)

3 (60%)

4 (80%)

2 (40%)

1 (20%)

3 (60%)

4 (80%)

3 (60%)

3 (60%)

1 (25%)

1
2
3
4-6

1 (20%)
1 (20%)

1 (20%)
1 (20%)

3 (75%)

4 (100%)

1 (14%)
6 (86%)

7 (100%)

1 (20%)

1 (20%)

1
2
3
4-7

2 (40%)
2 (40%)

2 (40%)
2 (40%)

1 (25%)
3 (75%)

4 (100%)

7 (100%)

7 (100%)

3 (60%)

1 (20%)

1
2
3
4
5-9
10+
0

1 ((20%)
1 (20%)
-

2 (40%)

2 (40%)

1 (25%)
3 (75%)

Not tested

5 (100%)

Not tested

Part 1
Situated understanding of LD
numeric problem-rephraseMCQ

Situated use of LD Vocabulary

(2)

Situated, scaffolded
understanding of LD numeric
problem-start step

Scaffolded solving of multistep LD numeric problem;

interpreting answers, using
language

Part 3

Part 4

Part 5

Scaffoldedchecking of LD
solution; interpreting answers,
using language

Unscaffolded solving of multistep real-life problem needing

LD and other operations with 2
and 3 digit divisors;
interpreting answers, using
language

Unscaffolded- solving of multistep real-life problem with 2-3

digit divisors and decimal
numbers, needing LD and other
operations; interpreting
answers, using LD language

4 (100%)
4 (100%)

1 (25%)
7 (100%)

7 (100%)

3 (75%)

7 (100%)

7 (100%)

4 (100%)

1 (14%)

3 (75%)
4 (100%)

4 (100%)

7 (100%)

1 (25%)

6 (86%)

4 (100%)

1 (14%)
1 (14%)
5 (71%)
Not tested

7 (100%)

7 (100%)

3 (75%)
1-4
1 (25%)
7 (100%)
5-9
*Note: See Appendix B for examples of ordered items, design specifications, and statistics on item difficulty and discrimination in Years 1and 2.

39
Figure 1. Case Study of Class Exposed to PALD-Trained Teachers for Two Years: Changes in Mastery of Long
Division at Three Developmental Levels
Developmental Level 1: Change from Year 1-Year 2 on Long Division Domain (N=5)
long division raw total score YEAR 1

long division raw total score YEAR 2

Frequency

Frequency

Mean =11.70
Std. Dev. =7.95
N =5

Mean =13.00
Std. Dev. =8.456
N =5

5.00

10.00

15.00

20.00

25.00

30.00

5.00

10.00

long division raw total score YEAR 1

15.00

20.00

25.00

30.00

long division raw total score YEAR 2

Developmental Level 2: Change from Year 1-Year 2 on Long Division Domain (N=4)
long division raw total score YEAR 1

long division raw total score YEAR 2

2.0

1.5

Frequency

Frequency

1.0

0.5

Mean =27.00
Std. Dev. =9.018
N =4
0.0

Mean =34.50
Std. Dev. =1.732
N =4
0

10.00

15.00

20.00

25.00

30.00

35.00

32.00

long division raw total score YEAR 1

34.00

36.00

38.00

long division raw total score YEAR 2

Developmental Level 3: Change from Year 1-Year 2 on Long Division Domain (N=7)
long division raw total score YEAR 2

2.0

1.5

Frequency

Frequency

long division raw total score YEAR 1

1.0

0.5

Mean =44.43
Std. Dev. =1.813
N =7

Mean =30.43
Std. Dev. =4.65
N =7

0.0

0
20.00

22.50

25.00

27.50

30.00

long division raw total score YEAR 1

32.50

42.00

44.00

46.00

long division raw total score YEAR 2

48.00

40
Figure 2. Cognitive Processes of a Learner at Developmental Level 1: Pathways to Mastering Long Division

41

Figure 3. Cognitive Processes of a Learner at Developmental Level 3: Pathways to Mastering Long Division

42
Appendix A
Graphic Representation of the Proximal Assessment for Learner Diagnosis (PALD) Model: How Teachers Work in
the Classroom

I. Goal-Setting
Specify: Short-term outcome in terms of culminating
task(s) to be mastered by learners and embedded domain of
concepts/skills
REVISE

Specify: Long-term
instructional goals to begin a new
unit (Possible Sources: State/District
Standards and Curriculum Maps in
Subject Areas)

II. PALD Planning

Design instruction and assessment:

Administer Assessments for

Summative Decisions

REVISE

IV. End of Unit

Develop lesson plans linked to domain

Design instructional strategies linked to domain
Develop developmentally-diagnostic assessments linked to
domain

III. PALD
Instruct (new concepts)

Begin with and Embed Proximal Assessments for

Learner Diagnosis (PALD) to Detect Learning Gaps

Administer Assessments

Conduct Error Analysis

Mediate and Give

Differentiated Feedback

Craft Intensive Practice

Opportunities

Make Formative Decisions

Are learners ready for new concepts/unit?

43

Appendix B
Developmentally-Ordered Diagnostic Assessment in Long Division (LD): Shifts in Item Difficulty and
Discrimination Statistics
Long Division (LD) Domain:
Task Examples

Task Specifications, Order, and

Embedded Skills
and/or Concepts from LD Domain

Part 1. Look at the problem below to

answer Questions 1-6.

Question or probe situated in a numeric

LD problem that is presented first with 1-3
digit divisors. Problem is not solved.
(Learners) interpret/show understanding of
LD operation by restating what the
problem is asking. Structured-response
item format (scaffolded).

5 175

Year 1:
Difficulty index (Item
Discrimination in
parenthesis)

Year 2:
Difficulty index (Item
Discrimination in
parenthesis)

.75 (+0.26)

.94 (+.43)

Not tested

.75 (+.63)

.88 (+.61)

.87 (+.15)

.81 (+.35)

.94 (+.46)

.63 (+.83)

.75 (.85)

1. The problem is telling me to:

Intended difficulty: Low
a.
b.
c.
d.

Multiply 5 times 175

Find out how many groups
of 5 are in 175
Divide 5 into 175 parts
Add 5 and 175

5 175
2. Say in another way what the
problem is asking. _________

3. The 5 in the division problem is

called the
a.
b.
c.
d.

divisor
dividend
quotient
multiple

5. To start to solve the problem shown,

I would ask which questions?
a.

b.
c.

d.

How many times does 5 fit

into 1? And if not, how
many times does 5 fit into
17?
What is 5 times 1? And if
not, what is 5 times 17?
What is 5 divided by 17?
And if not, what is 5 divided
by 175?
Other (Explain): _________

8. Now do the problem, showing all

your work in the spaces marked below.

Question or probe situated in a numeric

LD problem with 1-3 digit divisors that is
presented first. Problem is not solved.
(Learners) interpret/show understanding of
LD operation by restating what the
problem is asking. Open-ended item
format (un-scaffolded)
Intended difficulty: Mid-Level
Question or probe situated in a numeric
LD problem with 1-3 digit divisors that is
presented first. Problem is not solved.
(Learners) show LD vocabulary
knowledge in problem-solving context.
Structured-response item (scaffolded)
item.
Intended difficulty: Low
Learners show understanding of how to
start LD operation, with probe situated in
a given LD numeric problem that is
presented first. Problem is not solved.
Structured-response (scaffolded) item
Intended difficulty: Low

Following preceding probes into

concept/operation understanding, learners
apply the multi-step procedure taught, to
solve a numeric LD problem with 1-3 digit
divisors and dividend larger than divisor.
Interpret the complete answer or parts of
answer. Open-ended, constructed
response item, with grouped scoring of
both parts together OR with analytic
rubric
Intended difficulty: Mid- to High Level

44

Long Division (LD) Domain:

Task Example

Task Specifications, Order, and

Embedded Skills and/or Concepts

Part 2.
For Questions 11 through 13, look at
this problem:

Question or probe situated in a numeric

LD problem that is presented first, with 13 digit divisors and dividend larger than
divisor. Problem is solved. Learners apply
LD vocabulary, concept and procedural
knowledge to check solution. Structuredresponse item (scaffolded).

4
5 23
20
3R

Year 1 Item Statistics:

Difficulty index (Item
Discrimination in
parenthesis)

Year 2 Item Statistics:

Difficulty index (Item
Discrimination in
parenthesis)

.88 (+.85)

.81 (+.91)

Intended difficulty: Mid-to High Level

11. Identify the following:

a. Quotient ______________
b. Divisor _______________
c. Dividend ______________
d. Remainder _____________
12. Which backwards operations
would you use to check the answer?
Choose one.
a.
5x4=
b.
5x4+3=
c.
5 x 23 =
d.
23 5 =
e.
Other _____________
13. Now use your chosen formula to
check the answers given. Is the
solution correct? Show your work.
Part 3.
Coconut cookies come in packets of 12.
I would like to give 1 cookie to each 5th
grader at Hillcrest Elementary. There
are 119 5th graders here. How many
packets of cookies should I buy so that
every 5th grader gets at least one
cookie? Will there be any cookies left
over? Set up all the operations you need
to solve the problem, solve it and
explain all the parts of your answer.
Show all your work in the spaces
provided.
The divisor in the problem is: _______
What does it stand for in the story
problem?________.
Part 4.
Mr. Generosa would like to buy snacks
for his students on an upcoming
fieldtrip. His budget is $25 and there
are 20 students in his class. If Mr.
Generosa wants to divide the money
equally among his students, then how
much would the snack for one student
cost? Set up the problem clearly and
show all your work to get the answer in
decimal form.

(Learners) apply multi-step LD procedure

to solve a real life problem with 1-3 digit
divisors and dividend larger than divisor.
In context, they use LD vocabulary,
concept knowledge, doing all other
operations needed to reach solution, and
interpret answer sensibly for problem,
completing all steps. Open ended-item,
with 5-7 point rubric. Interpretive items
have structured response format
(scaffolded).

.75 (+.94)

Not tested

.50 (+.69)

Intended difficulty: Mid- to High Level

(Learners) apply multi-step LD procedure

to solve a real life problem (1-3 digit
divisors and dividend larger than divisor)
with use of decimals in quotient, leaving
no remainder. They use vocabulary,
concept knowledge, doing all operations
needed to reach solution, interpret answer
sensibly for problem, and complete all
steps. Open ended-item, with 5-7 point
rubric. Interpretive items have structured
response format (scaffolded or
unscaffolded).
Intended difficulty: High

Class N=16

.75 (+.94)