Who are the Learning Disabled Is There a Future for A Cognitive Basis Evidence from MetaAnalyses and

1
Who are the Learning Disabled?Is There a Future
for A Cognitive Basis?Evidence from
Meta-Analyses and Longitudinal Research

• H. Lee Swanson
• University of California-Riverside
• Institute for Education Sciences
• June , 2009

2
Overview of Meta-Analyses

• 1. Meta-analyses of Cognitive and RTI Research
  (in process)with Brenda Arellano, Loan Tran and
  Tori Sanchez
• 2. Meta-Analysis of Adults with RD
• Funded by NIFL (with Ching-Ju (Rosie) Hsiehunder
  review
• 3. Meta-Analysis of Memory and RD
• (JLD,2009 with Xinhua Zheng and Olga Jerman)

3
Overview of Reading and Math Projects

• 4. Meta-Analysis of Correlational Data on
  Phonological Awareness, Rapid naming and Reading
• (Review of Educational Research, 2003)
• Key Collaborators Guy Trainin, Denise Necoechea
• 5. Meta-analysis of Discrepancy and
  Non-Discrepancy Poor Readers (School Psychology
  Review-2000)
• Key CollaboratorMaureen Hoskyn

4
Overview of Reading and Math Projects

• 6. CurrentMath Disabilities vs. RD, RD and
  MD2007 Review of Educational Research
• Key collaboratorsOlga Jerman, Georgia Dukas,
  Rebecca Gregg
• 7. Meta-Analysis of Experimental Intervention
  Research in LD (RER, 1998, JLD, 2000, 2003)
• Several collaborators
• Key collaborators- Maureen Hoskyn and Carole Lee

5
Issue 1. Progress Toward Operational Definitions

• Agree on Subtypes relevant to academic
  outcomes----Three currently
• Newdirections explore high order
  definitionsproblem solving, comprehension
• Explore Cognitive Basis for definitions

6
Assumption related to the definition

• 1. Not due to inadequate opportunity to learn,
  general intelligence, or to significant physical
  or emotional disorders, but to basic disorders in
  specific psychological processes (e.g.,
  remembering the association between sounds and
  letters).
• 2. Not due to poor instruction, but to specific
  psychological processing problems that have a
  neurological, constitutional, and/or biological
  base.
• 3. Not manifested in all aspects of learning.
  Such individuals psychological processing
  deficits depress only a limited aspect of
  academic behavior. For example, such individuals
  may suffer problems in word recognition, but not
  calculation.

7
How researchers generally operationalize SLD

• 1. There are two subtypes that have some
  consensus
• reading disabilities and mathematical
  disabilities. also consider comorbid group
• 2. These subtypes are defined by standardized
  (normed referenced) and reliable measures of
  intelligence and achievement. The most commonly
  used intelligence tests are from the Wechsler
  measures and common achievement tests that
  include measures of word recognition or
  arithmetic calculation (e.g., WIAT, WRAT, WRMT).
• 3. In general, individuals with IQ scores (e.g.,
  verbal) equal to or above a standard score of 85
  and reading subtest scores equal to or below the
  25th percentile and/or arithmetic subtest scores
  equal to or below the 25th percentile reflect two
  high incidence disorders within LD reading (word
  recognition), and arithmetic (computation,
  written work).
• 4. By far, the subtype that has received the most
  research attention is reading disabilities.

8

• Some issues in the area of cognition
• 1. Clouded by conflicting evidence on IQ and
  reading discrepancy research
• (logic that similarities in overt behavior
  reflect the same inefficiencies and/or of
  cognitive processesconsider MD and RD)
• 2. Clouded by previous cognitive intervention
  research (poor generalization to changes in
  academics)
• 3. Psychometric aspects of measures in question

9

• 4. Inadequate research framework which clouds
  interpretation of outcomes
• 5. Knowing cognitive deficits does not indicate
  teaching directions
• 6. There are few analogs (low inference
  observation measures) linking cognitive
  performance to classroom performance

10
Perspective

• 1. Purpose of assessing cognition is to explain
  the why and predict how individual
  differences account for treatment outcomes
• 2. Purpose of an instructional approach (e.g.,
  RTI) is to monitor the intensity of intervention
  (instruction) and make systematic changes as a
  function of overt performance
• Pt----the approaches are complementary -
• Pt-The study of cognition has the potential to
  outline constraints in learning when individual
  differences cannot be explained as a function of
  best instructional practice
• Ptinstruction accounts for less than 20 of the
  variance in effect sizes (Swanson, 1999
  Simmerman Swanson, 2001)

11
Table 2. Regression model predicting effect size
as a function of methods composite score, age,
and instructional components
12

• Table 2
• Predictions of Year 3 Problem Solving Accuracy
  Based on Wave 3 Math Calculation,
• Problem Solving Knowledge and Wave 1 Fluid
  Intelligence, Reading and Cognitive
  Variables
• Model 5 B SE ß t
• Wave 3 Predictors
• Problem Solving Knowledge 0.25 0.11 0.12 2.13
• Calculation 0.30 0.08 0.27 3.42
• Wave 1 Predictors
• Fluid Intelligence (Raven) 0.13 0.04 0.16 2.85
  
• Reading 0.12 0.12 0.12 1.00
• Phon. Know. -0.01 0.10 0.10 -0.09
• Fluency 0.02 0.07 0.007 0.33
• Speed -0.004 0.06 -0.004 -0.06
• Inhibition 0.09 0.06 0.07 1.60
• Age -0.15 0.06 -0.16 -2.39
• Sketchpad 0.15 0.04 0.14 3.23

13
Math Calculation
14
Reading Composite
15
Phonological Processing
16
Word Problems
17
A Focus on the Instructional Side of LD
18
Issue 2 Determine Meaningful Outcomes

• 1. Control group needs to include significant
  instructional moderators (e.g.,DRP, overlap with
  treatment)
• 2. Determine the role definitional moderators

19
Why Do A Meta-Analysis to address these questions?

• 1. Evidence BasedPattern across several studies
  vs. single study---vs. overstated or understated
  information
• 2. Influence of sample (age, IQ, Discrepancy) and
  intervention parameters (time,responsiveness vs.
  resistance to instruction, components of
  instruction) on outcomes.
• 3. Theory Testing---identify the core
  problem---area most resistant to intervention
• 4. Allows for Replication

20
Interpretation

• For the purpose of discussion, Cohens (1988)
  distinctions on the magnitude of the effect size
  will be used.
• .20 is a small size
• .60 is a moderate size
• .80 is a large effect size

21
Table 1 NRP
22
Can we ignore cognition by focusing primarily on
evidence-based instruction ?What do we know
related to evidence based intervention and where
should we go?1. Meta-Analysis of Experimental
Interventions and LD (e.g., RER, 1989, JLD
2001).2. Meta-analysis of Dynamic Assessment
(e.g., RER, 2001)3. Meta-analysis of RTI
research (in progress)
23
Criteria

• Computer search, dissertations, state department
  reports-? 3000 manuscripts
• . Control group, average intelligence, minimum of
  3 sessions, ES can be calculated.
• Final 180 group design (K1,537) and 85 single
  subject design studies (K793)

24
Bottom line for evidence based studies

• 1. Mean ES between LD in control and TRT .56
• 2. Mean ES between LD (Exp. TRT) and NLD in .97
• 3. Majority of Studies measure Reading
• 4. Several variables significantly moderate
  treatment outcomes (IQ Reading, teacher
  effects, components overlap, standardized vs.
  experimental measures, ratings on internal and
  external validity)
• 5. Combined Strategy and Direct instruction most
  robust procedure

25
9

22

.41

.33

.23

.46

.05


6b. Experimental

21

49

.59

.42

.34

.51

.04


7.

Writing

19

67

.84

.63

.54

.72

.05


7a. Standardized

3

7

.37

.36

.14

.58

.
II


7b. Experimental

16

60

.80

.68

.59

. 78

.04


8.

Vocabulary

11

20

.79
.78

.66

.89

.05



26
(No Transcript)
27
TABLE 25. Mean Effect Sizes on Instructional
Components Comparing LD in Treatment Conditions
to NLD Participants
28
Issue 3 Determine the moderating role of IQ
29
What about Bob (IQ) ?

• 1. Does IQ relate to treatment outcomes ?
• Rephrase the questionif IQ is left out of the
  definition will it influence treatment outcomes?

30
Instructional Outcomes as A Function of IQ and
Reading Level
31
(No Transcript)
32
Bottom Line ON IQ

• 1. LD in Exp. Condition vs. average ES.69 for
  IQRD information
• ES1.41 for no IQ RD information
• 2. LD in Exp. vs. LD in Control
• ES.63 for IQRD information
• ES.82 for no IQ RD information
• ES.60 for IQRDMath information

33
Mixed Regression Modeling for Predicting
Estimates of Effect Size in Cognitive Processing
34
A speculation

• Based on studies that include optimal
  instructional conditionsthe mean effect size one
  could expect comparing LD with nonLD is (tier 2
  or 3)----
• D-R-P (.78)
• Systematic Probing (.73)
• Peer Mediation (.52)
• Strategy Cuing (.74)
• Mean ES.69 under evidence based
  instruction---which may varying depending on the
  entry of new data--

35
Issue 4Develop Standardized Measures related to
Dynamic Assessment

• Can we detect LD early with DA procedures?---longi
  tudinal research
• DA of Cognitive and/or Academic?

36
Synthesis of Experimental on Dynamic Assessment
(RER, 2001)

• Criteria for Selection
• 1. Published Refereed Journal
• 2. Control group comparison (between and within
  comparisons) for DA vs. static or traditional
  measurement (no feedback)
• 3. 30 articles from 303 potential (majority
  eliminated because ES could not be calculated,
  duplicate data) articles analyzed

37
Questions

• Is new information gained by DA procedures
  relative to traditional assessment?
• Are some groups of children more responsive then
  others?
• Which DA procedures yield the highest outcomes
  (relative to traditional assessment )?

38
(No Transcript)
39
Results and Implications for LD---DA vs.
traditional

• 1. Lower effect sizes emerge for LD relative to
  other categories of children
• 2. Largest ESs occur for underachievers
• 3. Testing limits (e.g., scaffolding---various
  cuing procedures) and general strategies (general
  feedback, modeling strategies) yielded higher
  outcomes than test-train-test models
• Implication---LD sample performance as a function
  of DA is hard to change relative to other groups--

40
Issue 5 determine if RTI studies can
change risk factor of children already with
serious risk factors (beyond what psychometric
studies can provide)----is there better
explanatory power knowing general areas of
cognition?

• IS RTI itself a wait and fail situation???

41
Meta-analytic look at RTI findings

• Criteria for Selection
• 1. Published Study (1985-2008)-
• 2. Divided sample into
• responders and nonresponders
• 3. Focus on reading-Elementary
• 4. Reported Pretest Scores by Responders and
  Nonresponders
• 5. Reported Standardized scores
• 6. Allow for calculation of ES
• Only 9 studies met criteria (119 ES)

42
Issues facing RTI and how a Meta-analysis can help

• 1. No (or few) systematic control studies (none
  meeting the gold standard) comparing RTI with a
  competing model of classification
• 2. RTI is a function of instruction (as well as
  teachers), and because there is no standardized
  protocol for instructionhow well can people
  generalize from findings to classify child at
  risk across school districts?
• 3. No consensus on definition of what resistance
  to instruction should be (slope of 0 or .25 or
  benchmarks?)---Is the issue really intercept
  level and not change (slope)?

43
Group Design RTI Studies (Responder vs.
nonresponder in the same evidence-based
intervention)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
Tentative Conclusions (RTI Studies)

• 1. Pretest differences for some children
  seriously at risk remain stableand a source for
  determining LD
• 2. Instruction is not robust enough to ignore
  individual differences in achievement and
  perhaps cognition

48
Conclusion on InterventionsWho are the SLD ?

• Children who yield low outcomes under optimal
  instructional conditions (components that
  significantly and positively influence effect
  sizes
• Those optimal instructional components that
  predicted treatment outcomes---Drill-repetition-sk
  ills, strategy training and small interactive
  groups
• Definition does influence outcomes-IQ and Reading
  Scores in combination are not irrelevant to
  instructional outcomes (at least from this data
  set). Average IQ and low Reading group (lt 25th
  percentile) appears to be one subgroup most at
  risk in terms of the magnitude of outcomes.

49
The Assessment Side of LD

• What Cognitive Variables are Important in
  Assessing RD and MD in children?
• Do risk factors related to Cognition go away in
  adulthood?

50
Issue 6 What are the important cognitive
processes to consider in in children and
adults with LD ?

• What are the common cognitive deficits among
  subgroups?
• What are the non-overlapping cognitive deficits
  among subgroups?
• What deficient cognitive processes operate
  independent of classification variables--

51
Math disabilities Meta-analysis of published
literature
52
Research and Policy Question

• 1. Are cognitive deficits comparable between RD
  and MD children?
• 2. Does the identification of cognitive processes
  help in the classification (does it matter) ?

53
Selection criteria

• 1. Pool in excess of 800 articles
• 2. 85 articles with defined control groups
  (needed at least a nondisabled control group)
• 3. Standardized math,reading, and IQ scores
• 4. 28 studies met full inclusion criteria

54
Table 2 Psychological and Demographic
Information on Participants
55
Table 2 Continued
56
Table 2 Continued
57
Table 3 Weighted
Effect Sizes, Standard Error, Confidence
Intervals and Homogeneity of Categories for
Comparisons between MD and non math disabled
(MD/NMD), MD and reading disabled (MD/RD), and MD
and RDMD (CMOR) (corrected for outliers).
58
Table 3 Continued
59
Table 3 Continued
60
Table 3 Continued
61
What About Applied Cognition (Memory)JLD-2009

• 1. Published studies1970-2008
• 2. Defined RD and CA matched NONRD sample by
  Standardized scores
• 3. Outcomes on at least one STM or WM measure
  (operationally defined)
• 4. 88 studies, weighted ES-.89, STM-.65,
  WM-.67
• 5. Low IQLow Reading ES (RD vs. NRD)-.49,High
  IQlow reading-.85
• 6. 52 of between study variance explained by
  Memory

62

• W
• Effect Size as a Function of Categorical
  Variables When Compared to Chronological Age and
  IQ Matched
• CCategory Number of Studies
  M SD K Weighted Effect Size
  95 CI for effect size
• Lower Upper
• SShort-Term Memory
• 1. Phonological 7 -0.83 1.15 22 -0.39 -0.50 -0.29
  
• 2. Pictures 17 -0.90 1.13 53 -0.57 -0.65 -0.49
• 3. Words 25 -0.50 0.66 76 -0.55 -0.61 -0.48
• 4. Digits 11 -1.49 2.2 55 -0.63 -0.69 -0.56
• 5. Letters 4 -1.06 0.52 13 -1.10 -1.24 -0.95
• DDual Task-Trade-off-reorder
• 6. Backwards 16 -0.70 0.45 59 -0.69 -0.74 -0.63
• 7. Preload 3 -0.53 0.27 7 -0.49 -0.73 -0.26

63

• NNumber of Studies M SD K
  Weighted Effect Size
  95 CI for effect size
• Lower Upper
• WWorking Memory-D C format
• 9 Counting 10 -0.88 0.55 32 -0.78 -0.84 -0.73
• 1Listen/Sentence19 -1.51 1.21 57 -0.84 -0.89 -0.79
  
• 1 Visual- Matrix 26 -0.69 0.63 72 -0.80 -0.86 -0.7
  4
• 1 Complex Visual.6 -0.52 0.17 20 -0.48 -0.57 -0.39
  
• 1 Semantic Assoc.10 -0.81 0.44 31 -0.37 -0.44 -0.3
  0
• 1 Digit/Sentence10 -1.47 2.25 24 -0.58 -0.68 -0.48
  
• Story Retelling 4 -0.80 0.7 9 -0.37 -0.50 -0.24
  
• 1Phonol/Rhyming 7 0.32 13 -0.61 -0.74 -0.4
  9 -0.72
• D CDaneman and Carpenter task format

64
Rapid Naming, Phonological Awareness and Reading
65
Big Question

• Is Phonological Awareness the most important
  variable in predicting reading accuracy? or is a
  more comprehensive cognitive battery called for?

66
Research Questions

• 1.What is the correlational evidence on the
  relationships between phonological awareness,
  rapid naming speed, and sight word recognition?
• 2. Do other processes play an important role?
• 3. Are the correlations between RAN and PA
  independentare they sensitive to age?

67
Selection Criteria

• Dates 1966-2001-Include PA, RAN, and reading (138
  studies)
• 35 Studies Meeting Selection Parameters (report
  SD, complete intercorrelations)
• Correlations (K2,257)

68
Table 4Estimated Intercorrelations Among
Cognitive Measures
69
Conclusions

• 1. Predictions of real word reading
• No clear advantages for PA and RAN when compared
  with other variables
• 2. Role of Chronological ageage did not appear
  to play a moderating role in the correlations
  between RAN and PA

70
Do process deficits go away with time? Adult
Outcomes

• 1. Pool in excess of 450 articles- samples gt 18
  yrs of age and reading scores
• 2. Articles with RD and defined control groups
  (needed at least a nondisabled control group)
• 3. Standardized reading and IQ scores
• 4. Reported measures independent of
  classification measures
• 5. 52 studies met full inclusion criteria

71
(No Transcript)
72
(No Transcript)
73
(No Transcript)
74
(No Transcript)
75
(No Transcript)
76
(No Transcript)
77
RD vs. Slow Learners More Alike than Different?
78
Big Question

• Is IQ completely irrelevant in separating various
  reading groups??

79
Research Questions

• 1. Is the phonological core deficit the only
  process that holds between the two groups?
• 2.Are the effect sizes moderated by Age and/or
  verbal IQ?
• Problemwhats low achiever (lt 96 on IQ and
  reading40th percentile)
• Whats RD (25th percentile in reading and verbal
  IQ gt 80---designated discrepancy)

80
Selection of Studies

• 1. 20year periodmust include comparison of
  Discrepancy and Non Discrepancy groups
• 2. Criterion measure was reading recognition
• 3 Must report Standardized Intelligence and
  Reading Measures
• 4. Published in a refereed journal-English
• 69 potential articles
• 19 met criteria-274 effect sizes
• Mean ES .21 (SD.65)

81
Table 1 Age and Psychometric Characteristics of
Children with RD and Low Achievers
82
Magnitude of Effect Size by Category of Dependent
Measure.
83
Conclusions

• Verbal IQ and Age moderate the overall level of
  cognitive performance.
• This conclusion is different than saying IQ is
  irrelevant

84
Overall Conclusions

• Who are the SLD?
• Children with average IQs (gt84) with reading
  and/or math scores below the 25th percentile
  whose academic performance outcomes remain below
  an ES of .70 (when compared to normal achieving
  peers) after intense exposure to optimal
  instructional conditions

85

• 1. There evidence to suggest that IQ (at least
  verbal IQ) should not be thrown out of the
  definition.
• 2. Two processes are critical (PA, WM) when
  determining the subtypes of disabilities.
• 3. Children at great risk for SLD are those
  exposed to optimal instructional conditions who
  are in the average range of intelligence and
  also experience processing inefficiencies in PA
  and WM.

86
Where to go from here?
87
Validating a Science Based Model of Learning
DisabilitiesSwanson, H.L. (2008) Neuroscience
and Response to Instruction (RTI) A
complementary Role In C. Reynolds E.
Fletcher-Janzen (Eds.) Neuropsychological
Perspectives on Learning Disabilities in the Era
of RTI Recommendation for Diagnosis and
Intervention. NY John Wiley Sons. Steps 1-3
88
Step 4
89
Step 5
90
Step 6
91
Step 7

• Step 7 Formulate a metatheory of learning
  disabilities by designating the parameters
  susceptible and not susceptible to instruction
• a.If anomalous data occur, return to Step 1.b.
• b. If additional data confirm theory, broaden
  context (e.g., determine influence of
  non-cognitive classroom variables on learning).