English Language Proficiency Tests, One Dimension or Many?: Yoonsun Lee Director of Assessment and...

12
English Language Proficiency Tests, One Dimension or Many?: Yoonsun Lee Director of Assessment and Psychometrics Office of Superintendent of Public Instruction

Transcript of English Language Proficiency Tests, One Dimension or Many?: Yoonsun Lee Director of Assessment and...

English Language Proficiency Tests, One Dimension or Many?:

Yoonsun Lee

Director of Assessment and Psychometrics

Office of Superintendent of Public Instruction

ELPT Requirements Under NCLB States are required to:- Implement ELD standards- Implement ELP tests that assess skills in

listening, speaking, reading, and writing with an added comprehension measure

- Administer ELP tests annually in grades K-12

- Align ELP tests with academic content standards

- Meet AMAO Title III objectives

More on Title III Measurement Demands AMAO I requires setting target growth

rates in English language proficiency status across years

AMAO II requires setting targets for attaining full English language proficiency across years

States attracted to ELP tests that implement vertical scales

Construct Validity Issue Does it make sense to hypothesize that

English language proficiency test is unidimensional? Or, is it multidimensional with four different domains (reading, writing, speaking, and listening)?

Washington Language Proficiency Test-II (WLPT-II) Developed in 2006 Used Stanford English Language

Proficiency Test (SELP) and added augmented items developed by Washington teachers

Four grade spans (K-2, 3-5, 6-8, & 9-12) Four subtests (Reading, Writing, Listening

& Speaking)

WLPT-II Test Specifications

Grade Span

Reading Writing ListeningSpeakin

gTotal

Number

Primary(K-2)

21MC15MC8CR

20MC 17CR81

(112pts)

Elementary(3-5)

24MC20MC2CR

20MC 17CR83

(110pts)

Middle(6-8)

28MC24MC2CR

20MC 17CR91

(118pts)

High(9-12)

31MC24MC2CR

20MC 17CR94

(121pts)

MC: Multiple choice CR: Constructed response

Confirmatory Factor Analysis- 2006 WLPT-II- Sample: Approximately 15,000 students

included in each grade span- EQS (Bentler, 1995)- Four models were examined

Models 1 & 2 E1

E4

E5

E9

E10

E12

E13

E16

Rdg cluster 1

Rdg cluster 4

Wri cluster 1

Wri cluster 5

Lis cluster 1

Lis cluster 3

Spe cluster 1

Spe cluster 4

Language Proficiency

. .

. .

. .

. .

E1

E4

E5

E9

E10

E12

E13

E16

Rdg cluster 1

Rdg cluster 4

Wri cluster 1

Wri cluster 5

Lis cluster 1

Lis cluster 3

Spe cluster 1

Spe cluster 4

Language Proficiency

. .

. .

. .

. .

Models 3 & 4E1

E4

E5

E9

E10

E12

E13

E16

.

Rdg cluster 1

Rdg cluster 4

Wri cluster 1

Wri cluster 5

Lis cluster 1

Lis cluster 3

Spe cluster 1

Spe cluster 4

Reading

Writing

Listening

Speaking

Language Proficiency

E1

E4

E5

E9

E10

E12

E13

E16

.Rdg cluster 1

Rdg cluster 4

Wri cluster 1

Wri cluster 5

Lis cluster 1

Lis cluster 3

Spe cluster 1

Spe cluster 4

Reading

Writing

Listening

Speaking

D1

D2D3

D4

. .

..

.

. .

. .

. .. .

Results (Primary level)Model df GFI CFI RMSEA

1 (single with no corr)

16896.2 189 0.67 0.75 1.33

2 (single with corr)

2445.4 142 0.95 0.97 0.05

3 (four factor)

5472.14 183 0.88 0.92 0.07

4 (second order)

6192.9 185 0.87 0.91 0.08

2

Results (Primary: K-2)- was examined to compare models.

Model 2 (Single Factor with errors correlated within subtest) produced a good fit to the data.

2

Results (Elementary, Middle, & High School)

- Same result was found in Elementary, Middle, & High School)

- Model 2 showed the best fit to the data (over 0.95 GFI and CFI and below 0.05 RMSEA)

- No significant evidence to threaten construct validity with adding augmented items to the existing language test