sun-joo cho allan cohen brian bottge

21
Detecting Intervention Effects Using a Multilevel Latent Transition Analysis with a Mixture IRT model Sun-Joo CHO Allan COHEN Brian BOTTGE

Upload: adina

Post on 02-Feb-2016

27 views

Category:

Documents


0 download

DESCRIPTION

Detecting Intervention Effects Using a Multilevel Latent Transition Analysis with a Mixture IRT model. Sun-Joo CHO Allan COHEN Brian BOTTGE. Context. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Sun-Joo CHO Allan COHEN Brian BOTTGE

Detecting Intervention Effects Using a Multilevel Latent

Transition Analysis with a Mixture IRT model

Sun-Joo CHO

Allan COHEN

Brian BOTTGE

Page 2: Sun-Joo CHO Allan COHEN Brian BOTTGE

2

Context

• If you are asked by national Department of Education to test the efficacy of a new instructional method which was designed to improve learning disabilities students’ performance. Students’ performance were measured in two time points.

• People may concern that • (1) How much progress do student make? • (2) What’s the intervention effects?

Page 3: Sun-Joo CHO Allan COHEN Brian BOTTGE

3

• (3) What does the progress depend on? (intervention, teacher or both)

• (4) How do children’s response process change over time?

• (5) Do student perform differently?

Page 4: Sun-Joo CHO Allan COHEN Brian BOTTGE

4

Measuring Change

• CTT for measuring change

• 1. repeated ANOVA or MANOVA: mean change of a total score as a manifest variable across time points

• However, the assumption of sphericity (homogeneous and all correlation between any pair of measures equal) is often violated.

Page 5: Sun-Joo CHO Allan COHEN Brian BOTTGE

5

LTA-LCM

• Latent Transition analysis with Latent Class Model

• LTA-LCM assumes that there is no variability on the latent trait within classes.

LCM

LTA-LCM

Page 6: Sun-Joo CHO Allan COHEN Brian BOTTGE

6

LTA-MixIRTM

• Latent transition model with a mixture response model

Mixture Model

LTA_MixIRTM

Page 7: Sun-Joo CHO Allan COHEN Brian BOTTGE

7

Multilevel LTA-MixIRTMMixture Proportion in teacher level latent class

Mixture Proportion in individual level latent class

Transition proportion IRT model

Page 8: Sun-Joo CHO Allan COHEN Brian BOTTGE

8

• Proportion of the population in group-level latent class

• Proportion of the population in individual-level latent class at Time 1

• Transition proportion for t=2,…,T

Page 9: Sun-Joo CHO Allan COHEN Brian BOTTGE

9

Page 10: Sun-Joo CHO Allan COHEN Brian BOTTGE

10

Estimation

• Assumption: Item parameters are invariant across time and across group-level latent class (item parameters are estimated using post-test data)

• Anchor items (scale comparability among latent class?)

• Mplus (TWOLEVEL MIXTURE)

Page 11: Sun-Joo CHO Allan COHEN Brian BOTTGE

11

Empirical Study Results

• 310 students nested under 49 teachers answer 20 items at both pre-test and post-test occasion

Page 12: Sun-Joo CHO Allan COHEN Brian BOTTGE

12

• (1) Multilevel 1PL

ICC= 0.117 (pre-), 0.406 (post-)

(2) Detection of Latent Classes

A. Only one latent class at pre-test (very few students answer correctly)

B. Two latent class at post-test for student-level

C. Two latent class at post-test for teacher-level after the latent class at student level were determined

Page 13: Sun-Joo CHO Allan COHEN Brian BOTTGE

13

Page 14: Sun-Joo CHO Allan COHEN Brian BOTTGE

14

Page 15: Sun-Joo CHO Allan COHEN Brian BOTTGE

15

Page 16: Sun-Joo CHO Allan COHEN Brian BOTTGE

16

• Comparison between multilevel LTA-MixIRTM and LTA-MixIRTM: students were clustered into four transition patterns in multilevel LTA MixIRTM whereas there was two in the LTA MixIRTM.

Page 17: Sun-Joo CHO Allan COHEN Brian BOTTGE

17

Do students engage in problem solving differently?

Page 18: Sun-Joo CHO Allan COHEN Brian BOTTGE

18

How do students’ response change over time?

• On the post-test, 27.4% of the students were classified into student-level Class 2 (pattern 112 and 212). Higher transition proportion of pattern 212 indicate intervention effect.

Page 19: Sun-Joo CHO Allan COHEN Brian BOTTGE

19

• Pattern 212 has larger mean and large variance, indicating much progress.

How much progress? Teachers/intervention?

Page 20: Sun-Joo CHO Allan COHEN Brian BOTTGE

20

• In, Pattern 111 (pho=0.856), students tended to retain their relative positions.

• In Pattern 211 (pho=0.004), little variability for these student on post-test.

Page 21: Sun-Joo CHO Allan COHEN Brian BOTTGE

21

• Questions:

• 1. How about item parameter recovery?

• 2. Anchor item is necessary?

• 3. Achilles’ heel in mixture model