Title: Validity of Alternative Approaches to the Identification of LD: Operationalizing Unexpected Underach
1Validity of Alternative Approaches to the
Identification of LD Operationalizing
Unexpected Underachievement
- Jack M. Fletcher, Ph.D. Carolyn Denton
- U of Texas- Houston U of Texas- Austin
- CARS.uth.tmc.edu
- Jack.Fletcher_at_uth.tmc.edu
Responsiveness-to-Intervention Symposium December
4-5, 2003 Kansas City, Missouri The National
Research Center on Learning Disabilities, a
collaborative project of staff at Vanderbilt
University and the University of Kansas,
sponsored this two-day symposium focusing on
responsiveness-to-intervention (RTI) issues. The
symposium was made possible by the support of the
U.S. Department of Education Office of Special
Education Programs. Renee Bradley, Project
Officer. Opinions expressed herein are those of
the authors and do not necessarily represent the
position of the U.S. Department of
Education. When citing materials presented
during the symposium, please use the following
Fletcher, J. M., Denton, C. (2003, December).
Validity of alternative approaches to the
identification of LD Operationalizing unexpected
underachievement. Paper presented at the National
Research Center on Learning Disabilities
Responsiveness-to-Intervention Symposium, Kansas
City, MO.
2Learning Disabilities is a Construct
- Essential aspect of construct is unexpected
underachievement - Constructs do not exist independently of how they
are measured all measures are imperfect
indicators of constructs (latent variables) - Measurement depends on definition
- Definitions and the resultant identification
criteria derive from classifications - All classifications are hypotheses that must be
tested
3Learning Disabilities is a Construct
- Does the identification model that is derived
from the classification identify a unique
subgroup of underachievers? - Children may be validly identified even when the
underlying classification does not yield a unique
subgroup is the classification valid? - Validity of classifications only be tested on
variables not used to form the classification-
cognitive skills, response to intervention,
neuroimaging, genetics - Validity assumes reliability
4LD is a Valid Classification
- Learning disabilities are real! Stands up across
definitional variation (doesnt help identify
individuals) - Children and adults with different forms of
LD can be reliably and validly differentiated
from each other, typical achievers, and other
disabilities on cognitive correlates, response to
intervention, neural correlates, and heritability - What happens when we apply these criteria to
different classifications?
5Alternative Classification Models
- 1. Different forms of discrepancy VIQ, PIQ,
Listening comprehension- reading comprehension - 2. Low achievement
- 3. Intra- individual differences
- 4. Response to intervention
- 5. Hybrid models
6Alternative Discrepancy Models Discrepant vs.
LA(Fletcher et al., 1994)
80 impaired under all definitions difference is
in people above the low achievement cut point,
but below the regression cut point
7(No Transcript)
8Low Achievement
- Designate a cut point on the achievement
dimension - Strengths Strong validity, linked to
intervention, easy to implement - Weaknesses Cut point, does not measure the
underlying construct (cant differentiate
subgroups of poor readers when the cause is known
to be related to emotional difficulty, economic
disadvantage, and inadequate instruction) - Necessary but not sufficient
9Intra- Individual Differences Model
-
- while IQ tests do not measure or predict a
students response to instruction, measures of
neuropsychological functioning and information
processing could be included in evaluation
protocols in ways that document the areas of
strength and vulnerability needed to make
informed decisions about eligibility for
services, or more importantly, what services are
needed.
10Intra- Individual Differences Model
- An essential characteristic of SLD is
failure to achieve at a level of expected
performance based upon the students other
abilities (NCLD, 2002, p. 4). - Another version of a 2 test (or more)
discrepancy?
11Biobehavioral Systems Model
12Problems
- Processing subtypes, patterns weakly related to
intervention outcomes (if at all) - Not sure of what processes to measure outside
word recognition - What about non- MR children with flatter
profiles? (Biased towards milder impairments
because severity is correlated with shape) - Perpetuates assessment model (test and treat)
that has not been effective in enhancing outcomes - Difficult to scale (send all school psychologists
to NP school?) - But strong validity at achievement level (What
information is added outside variation in
achievement domains?)
13Response to Intervention
- Serial curriculum- based assessments of learning
in relation to an intervention - Identification is more reliable than when based
on a single assessment, even if the intercept is
used - As one criterion, student may be LD if they do
not respond to instruction that works with most
other students (I.e., unexpected
underachievement) - Identifies a unique subgroup of underachievers
that reflects an underlying classification that
can be validated (Al- Otaiba Fuchs, 2002
Vellutino et al., 2003)
14What percentage of G1 children dont respond
adequately to quality intervention?(Mathes et
al., 2003)
- Primary only 14/90 16 (3 of school
- population)
- Primary Secondary
- Proactive 1/82 lt 1 (lt .2 of school
population) - Responsive 7/83 8 (lt1.5 of school
population) - (Woodcock Basic Reading lt 30th percentile)
15Severe RD before and after 8 week intervention
(7- 17 years old) Simos et al. (2002)
16Not At Risk
17At Risk Responders
18At Risk Nonresponders
19What about the Nonresponders (Grade 2 and 3)?
Wave 1 Wave 2
Wave 3 Wave 4 Round 1
Phono-Graphix Read Naturally 8 weeks
8 weeks 8 weeks Round 2
Baseline Phono-Graphix Read
Naturally 8 weeks 8 weeks
8 weeks
20Woodcock-Johnson III Basic Skills Standard Scores
21Gray Oral Reading Test Fluency Standard Scores
22Gray Oral Reading Test Comprehension
23(No Transcript)
24Problems
- What is the best cut point? (index to benchmarks,
growth, norm referenced scores probably makes
little difference if measurement error taken into
account dont make it a formula) - Scaling and implementation problem is not with
the research base, but with its scaling (so what
else is new? 30 years later we still cant do IQ
discrepancy correctly) - Necessary but not sufficient needs additional
criteria
25OSEP LD Summit Hybrid Model
- 1. Low Achievement
- 2. Apply the Exclusions
- 3. Evaluate Response to Intervention
- (maintain flexibility of the interdisciplinary
team) - Unlike other alternative models, may yield a
valid classification of unexpected underachievers