1. 1 The Course of Reading Disability in First Grade: Latent Class Trajectories and Early Predictors Don Compton, Lynn Fuchs, and Doug Fuchs

2. 2 Criticisms of Current Learning Disabilities Definition Too many children are inappropriately identified Many children are classified as LD without participating in effective reading instruction in the regular classroom Too costly

3. 3 Criticisms of IQ-Achievement Discrepancy IQ tests do not necessarily measure intelligence IQ and academic achievement are not independent of each other In the case of word reading skill deficits, IQ- achievement discrepant poor readers are more alike than different from IQ-achievement consistent poor readers Children must fail before they can be identified with a learning disability

4. 4 What is Meant by an RTI Model? RTI refers to an individual, comprehensive student-centered assessment model. RtI is sometimes referred to as a problem-solving model. RtI models focus on applying a problem solving framework to identify and address the student’s difficulties using effective, efficient instruction and leading to improved achievement. RTI refers to an individual, comprehensive student-centered assessment model. RtI is sometimes referred to as a problem-solving model. RtI models focus on applying a problem solving framework to identify and address the student’s difficulties using effective, efficient instruction and leading to improved achievement.

5. 5 Typical RTI Procedure All children in a class, school, district are tested once in the fall to identify student at risk for long-term difficulties. The response of at-risk students to GE (Tier1) is monitored to determine whose needs are not met and therefore require more intensive tutoring (Tier 2). For at-risk students, research-validated Tier 2 tutoring is implemented. Student progress is monitored throughout intervention. Students are re-tested following intervention. Those who do not respond to the validated tutoring are identified As LD For multi-disciplinary team evaluation for possible disability certification and special education placement.

6. 6 Advantages of RTI Approach Provides assistance to needy children in timely fashion. It is NOT a wait-to-fail model. Helps ensure that the student’s poor academic performance is not due to poor instruction. Assessment data are collected to inform the teacher and improve instruction. Assessments and interventions are closely linked.

7. 7 Within RTI Identification Tier 2 tutoring is viewed as the “test” to which at-risk students respond to determine disability. Tier 2 tutoring is viewed as the “test” to which at-risk students respond to determine disability. That response needs to be measured and categorized as “responsive” (not LD) or “unresponsive” (LD) using an appropriate tool for such measurement. That response needs to be measured and categorized as “responsive” (not LD) or “unresponsive” (LD) using an appropriate tool for such measurement.

8. 8 RTI: Three Tiers Tier 1 Tier 1 − General education  Research-based program  Faithfully implemented  Works for vast majority of students  Screening for at-risk pupils, with weekly monitoring of at-risk response to general education Tier 2 Tier 2 − Small-group preventative tutoring − Weekly monitoring of at-risk response to tier 2 intervention Tier 3 Tier 3 − Special education

9. 9 Primary Prevention: School-/Classroom- Wide Systems for All Students, Staff, & Settings Secondary Prevention: Specialized Group Systems for Students with At-Risk Behavior Tertiary Prevention: Specialized Individualized Systems for Students with Intensive Needs ~80% of Students ~15% ~5% CONTINUUM OF SCHOOL-WIDE SUPPORT

10. 10 RTI Tier 2: Standardized Research-Based Preventative Treatment Tutoring Small groups (2-4) Small groups (2-4) 3-4 sessions per week (30-45 min per session) 3-4 sessions per week (30-45 min per session) Conducted by trained and supervised personnel (not the classroom teacher) Conducted by trained and supervised personnel (not the classroom teacher) In or out of classroom In or out of classroom 10-20 weeks 10-20 weeks

11. 11 What does Tier 2 look like? Hypothetical Case Studies

12. 12

13. 13

14. 14

15. 15 Purpose of the Study To explore: To explore: − Effects of multiple Tier 1 (classroom) and Tier 2 (pullout) instructional approaches on at-risk children’s reading growth in a 9-wk treatment period in fall of 1 st grade. − How responsiveness to the instructional approaches can be used to identify children as LD at the end of 1 st grade. − Effects of alternative methods of LD classification on prevalence and severity. − Can characteristic growth patterns of children who are either LD and not LD be identified for Tier 1 and Tier 2 instruction?

16. 16 Reading Study Sample 42 1 st -grade classes in 16 schools (8 Title) 42 1 st -grade classes in 16 schools (8 Title) Six lowest readers from each class on WIF and RLN, with teacher corroboration (252 low-study-entry children) Six lowest readers from each class on WIF and RLN, with teacher corroboration (252 low-study-entry children) Beginning 1 st grade, 6 children from each class rank ordered and, within class, split into 2 strata Beginning 1 st grade, 6 children from each class rank ordered and, within class, split into 2 strata Within each stratum within each class, randomly assigned to 3 groups (n = 84 per condition) Within each stratum within each class, randomly assigned to 3 groups (n = 84 per condition) − No tutoring (n=55 [65.5%] complete data at end grade 3) − Fall 1 st -grade tutoring (n=61 [72.6%] complete data at end grade 3) − Spring 1 st -grade tutoring, but only with inadequate slope/final intercept for fall 1 st grade (n=64 [76.2%] complete data at end grade 3) Three groups comparable demographically and on RLN, WIF, IQ, WRMT WID/WA, TOWRE SW/PD Three groups comparable demographically and on RLN, WIF, IQ, WRMT WID/WA, TOWRE SW/PD 18 weekly Word Identification Fluency measurements 18 weekly Word Identification Fluency measurements End of 3 rd grade, disability: <85 on latent variable of word reading, nonsense word reading, comprehension End of 3 rd grade, disability: <85 on latent variable of word reading, nonsense word reading, comprehension

17. 17 Evidence-Based Tutoring Tutoring Tutoring −Letter-Sound Recognition −Phonological awareness and decoding −Sight Words −Fluency Four Groups Four Groups − Fall Tutoring (n=61) − Spring Tutoring for Nonresponsive Children (n=32) − Spring No Tutoring for Responsive Children (n=32) − Controls (No Tutoring, n=55) Sessions Sessions −Conducted by research assistants −2-4 students per group −4 sessions/week −45 minutes/session −For a total of 36 sessions of tutoring

18. 18Questions Identify 1 st -grade growth trajectories characteristic of later disability versus ND Identify 1 st -grade growth trajectories characteristic of later disability versus ND Examine effects of 1 st -grade tutoring on trajectories Examine effects of 1 st -grade tutoring on trajectories Explore cognitive profiles associated with each latent class Explore cognitive profiles associated with each latent class

19. 19 General Model for Identifying Trajectory Classes

20. 20 Analysis Plan Conventional growth modeling to evaluate appropriateness of the hypothesized quadratic model Conventional growth modeling to evaluate appropriateness of the hypothesized quadratic model Multiple group growth mixture modeling with a distal latent factor (F, at end 3 rd grade in reading; end 2 nd grade in math) and beginning 1 st -grade covariates to identify disability and nondisability populations within each known group. Multiple group growth mixture modeling with a distal latent factor (F, at end 3 rd grade in reading; end 2 nd grade in math) and beginning 1 st -grade covariates to identify disability and nondisability populations within each known group. − Distal latent factor was regressed on the categorical latent variable (C), representing subpopulation CBM growth characteristics in 1 st grade. − Subpopulation variable (C) was regressed on the known class variable (CG). − Growth parameters (I, S, Q) and C were regressed onto the time-invariant covariates.

21. 21 Estimated Parameters of Interest Average latent class probabilities: likelihood each individual belongs to each class Average latent class probabilities: likelihood each individual belongs to each class Class-specific profiles: likelihood each individual in the class scores above/below criterion for disability on distal latent class indicator Class-specific profiles: likelihood each individual in the class scores above/below criterion for disability on distal latent class indicator Means/variances on Means/variances on − Growth parameters (I,S,Q) − Beginning 1 st -grade performance − Cognitive predictors − End-study performance as function of known class and disability/nondisability trajectory class − Class-specific probabilities for categorical latent variable as function of the covariates

22. 22 Data Analysis Growth model analyses with Mplus 4.0 Growth model analyses with Mplus 4.0 Model estimation used maximum likelihood estimator with robust standard errors Model estimation used maximum likelihood estimator with robust standard errors CBM data centered on initial assessment CBM data centered on initial assessment Mplus missing data module (maximum likelihood missing at random estimation procedures) Mplus missing data module (maximum likelihood missing at random estimation procedures) Estimated starting values derived from multiple group analysis of growth using only the CBM data Estimated starting values derived from multiple group analysis of growth using only the CBM data Covariates centered on grand means Covariates centered on grand means

23. 23 Results: Conventional Growth Modeling Word identification fluency (WIF) Word identification fluency (WIF) 18 weekly across fall and spring 18 weekly across fall and spring Quadratic model improved overall fit of model over linear model Quadratic model improved overall fit of model over linear model I: 14.20 words (SE=0.719; z = 19.74) I: 14.20 words (SE=0.719; z = 19.74) S: 1.80 words per week (SE=0.138; z = 13.09) S: 1.80 words per week (SE=0.138; z = 13.09) Q: -0.015 words 2 per week (SE=0.006; z = - 2.31) Q: -0.015 words 2 per week (SE=0.006; z = - 2.31)

24. 24 Results: Fall Tutoring

25. 25 Results: Spring Tutoring Necessary

26. 26 Results: Spring Tutoring Unnecessary

27. 27 Results: Control

28. 28 Results: Growth Mixture Modeling For each trajectory class, intercept and slope was significantly greater than zero and necessary for describing growth. For each trajectory class, intercept and slope was significantly greater than zero and necessary for describing growth. Quadratic term significantly different from zero only for Quadratic term significantly different from zero only for − Fall tutoring (z = -2.574) − Spring tutoring-necessary (z = 4.346)

29. 29 Average Probability of Latent Class Assignment and Class-Specific Profiles on the Distal Reading Latent Class Indicators Class-Specific Probabilities on Latent Class Indicators b Latent ClassLatent Class Probability a WRMT-R WID WRMT-R WA WRMT-R PC Fall Tutoring RD.964.022.501.005 Fall Tutoring NRD.995.954.999.833 Spring Tutoring Necessary RD.942.242.934.071 Spring Tutoring Necessary NRD.993.9851.000.941 Spring Tutoring Unnecessary RD.943.826.995.534 Spring Tutoring Unnecessary NRD.9271.000 Control RD.912.571.983.242 Control NRD.952.9841.000.937

30. 30 Results: Growth Mixture Modeling (across entire sample) Average latent class probability: Probability child is assigned to correct disability trajectory class within the known class:.912 to.995 (precise) Average latent class probability: Probability child is assigned to correct disability trajectory class within the known class:.912 to.995 (precise) Class-specific profiles on 3 rd -grade latent class indicators of disability (WID, WA, PC): Probability child in that class would score > 85 Class-specific profiles on 3 rd -grade latent class indicators of disability (WID, WA, PC): Probability child in that class would score > 85 − WA: Across disability groups, poor precision. − WID and PC: More consistently distinguished RD from ND. − For spring tutoring-unnecessary RD group, class-specific probabilities indicate this class does not have a characteristically RD profile. − For control RD group, high class probability of scoring normal on WID, but low class probability of scoring normal on PC. So, poor reading comprehension is the defining characteristic of untreated at- risk students.

31. 31 Estimated Multinomial Regression of Latent Class Variable on Covariates

32. 32 Estimated Multinomial Regression of Latent Class Variable on Covariates

33. 33 Plots represent estimated class-specific probability of class membership as function of one covariate, while keep other covariates constant Sound matching and vocabulary distinguished latent class membership, but only in control group. Sound matching and vocabulary distinguished latent class membership, but only in control group. Control students with lower sound matching scores have greater probability of being assigned to control RD class. Control students with lower sound matching scores have greater probability of being assigned to control RD class. Control students with higher vocabulary scores have greater probability of being assigned to control ND class. Control students with higher vocabulary scores have greater probability of being assigned to control ND class.

34. 34 Conclusions First-grade trajectory classes associated with 3 rd -grade disability status can be identified with high precision using WIF. So, WIF can be used for 1 st -grade progress monitoring within RTI, as an indicator of long-term RD status. First-grade trajectory classes associated with 3 rd -grade disability status can be identified with high precision using WIF. So, WIF can be used for 1 st -grade progress monitoring within RTI, as an indicator of long-term RD status. In control (untreated) group, RD and ND trajectory classes had same intercept, but vastly different slopes. So, slope can be used to index responsiveness. In control (untreated) group, RD and ND trajectory classes had same intercept, but vastly different slopes. So, slope can be used to index responsiveness. Only 2 classes had significant quadratic term. Only 2 classes had significant quadratic term. − For fall tutoring, growth decelerated across year. − For spring tutoring-necessary, growth accelerated across year.

35. 35 Conclusions 3 rd -grade WID and PC measures distinguished RD from ND; WA did not. 3 rd -grade WID and PC measures distinguished RD from ND; WA did not. Spring tutoring-unnecessary NRD was a relatively pure group of NRD students. So, using WIF in fall semester of 1 st grade to select children at-risk students may be efficient. Spring tutoring-unnecessary NRD was a relatively pure group of NRD students. So, using WIF in fall semester of 1 st grade to select children at-risk students may be efficient.

36. 36 Conclusions For control RD students, reading comprehension skill was defining characteristic. Interesting because 1 st -grade trajectory classes formed exclusively with WIF. Also, no way to distinguish control RD and NRD using intercept. For control RD students, reading comprehension skill was defining characteristic. Interesting because 1 st -grade trajectory classes formed exclusively with WIF. Also, no way to distinguish control RD and NRD using intercept. 1 st -grade cognitive predictors most useful for untreated students. For control students, low sound matching associated with RD; high vocabulary associated with NRD. 1 st -grade cognitive predictors most useful for untreated students. For control students, low sound matching associated with RD; high vocabulary associated with NRD. Within treated students, RTI (trajectory class) was what distinguished RD from NRD, effectively overriding initial individual differences on sound matching and vocabulary. Within treated students, RTI (trajectory class) was what distinguished RD from NRD, effectively overriding initial individual differences on sound matching and vocabulary.

37. 37 Thank You