less is more? an application of propensity score stratification to first-grade retention mieke goos,...

39
Less Is More? An Application of Propensity Score Stratification to First- Grade Retention Mieke Goos, Jan Van Damme, Patrick Onghena and Katja Petry SREE 2010

Upload: janis-elliott

Post on 27-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Less Is More? An Application of Propensity Score Stratification to First-Grade Retention

Mieke Goos, Jan Van Damme, Patrick Onghena and Katja Petry

SREE 2010

1. Introduction

• Starting point:

– Many young children struggle in elementary school

– Countries deal with these early problems in a different way

– Internationally frequently applied measure = grade retention

1. Introduction

• Grade retention in Flanders: – Relatively high rate

• For example: PISA 2003

1. Introduction

– Relatively high rate … especially in Grade 1 About 7% of Flemish children repeat Grade 1

– Socially approved by educators, policy makers and parents

→ being a grade retainee in Flanders has a different connotation than for example in the US (negative overtone)

– No formal rules regarding grade promotion (no national/state standardized test procedures)

→ retention decision = joint decision by teacher and parents

1. Introduction

• Research question:

– Is Grade 1 retention an effective practice or not?

• Focus of this study:

– children’s psychosocial growth throughout elementary school

2. Method

• Subjects: representative sample from the Flemish SiBO-project

– 3624 first-graders, of which 298 were retained

– 222 classes

– 121 schools

followed until Grade 6

2. Method

• Instruments

– Psychosocial growth: Teacher questionnaire

rated yearly by the teacher items on a 1 to 6 point Likert scale 7 subscales

Social skills• Popularity among classmates• Aggressive behavior • Hyperactive behavior• Asocial behavior

Dynamic-affective attitudes and skills • Independent participation• School well-being• Self-confidence

2. Method

• Instruments (continued)

– Propensity of repeating Grade 1

official records achievement tests Standard Progressive Matrices teacher questionnaire about the child parent questionnaire teacher questionnaire about teacher didactics school staff questionnaire

• 68 prior student characteristics• 59 prior class characteristics• 42 prior school characteristics

2. Method

• Analyses: 4-steps-procedure

– Step 1: identification of ‘true’ confounders of Grade 1 retention

prior student, class and school characteristics that are related to both treatment (i.e., Grade 1 retention)

and outcome (i.e., children’s individual psychosocial growth)

– Step 2: estimation of propensity scores based on these confounders

3-level logistic regression analysis (students – classes – schools)

2. Method

• Analyses: 4-steps-procedure (continued)

– Step 3: decile stratification

10 strata of equal size

– Step 4: estimation of average psychosocial effects

3-level curvilinear growth curve analyses (measurements – students – schools)

2. Method

• Analyses: 2 comparison strategies

– Same-grade approach = comparing retainees with their younger grade-mates

– Same-age approach = comparing retainees with their age-mates who were promoted to a higher grade

2. Method

A

Research year 1(age 7)

Research year 2(age 8)

Research year 3(age 9)

B

3rd grade

C2nd grade D

E

1st grade Grade retention

Grade retention

F

H

GGrade retention

Promotion

Promotion

Promotion

Promotion

Cohort 1

Cohort 2 (not in SiBO dataset)

A

Research year 1(age 7)

Research year 2(age 8)

Research year 3(age 9)

B

3rd grade

C2nd grade D

E

1st grade Grade retention

Grade retention

F

H

GGrade retention

Promotion

Promotion

Promotion

Promotion

Cohort 1

Cohort 2 (not in SiBO dataset)

2. Method

SAME-GRADE COMPARISON

A

Research year 1(age 7)

Research year 2(age 8)

Research year 3(age 9)

B

3rd grade

C2nd grade D

E

1st grade Grade retention

Grade retention

F

H

GGrade retention

Promotion

Promotion

Promotion

Promotion

Cohort 1

Cohort 2 (not in SiBO dataset)

2. Method

SAME-AGE COMPARISON

2. Method

• Analyses: 2 comparison strategies (continued)

– Why? ~ 2 different questions

How do Grade 1 repeaters, at the cost of one extra year of education, develop in comparison to younger children with whom they will eventually finish elementary school?

SAME-GRADE APPROACH

How would Grade 1 retainees have developed, had they been promoted to Grade 2 instead?

SAME-AGE APPROACH

3. Results

• Propensity scores

– based on 52 prior student characteristics

Promoted students

M = -4.47

Retained students

M = -0.12

3. Results

• Propensity score stratification

– Cut-offs for strata based on overlap

– Division into 10 strata of equal size

3. Results

– Within-stratum balance in propensity score

N M N M N M

1 233 -6,03 1 -6,32 234 -6,03

2 233 -5,57 1 -5,45 234 -5,57

3 233 -5,24 1 -5,23 234 -5,24

4 233 -4,93 1 -4,78 234 -4,93

5 232 -4,60 2 -4,70 234 -4,61

6 231 -4,30 3 -4,28 234 -4,30

7 234 -3,97 1 -3,81 235 -3,97

8 232 -3,58 3 -3,54 235 -3,58

9 230 -2,95 5 -2,90 235 -2,95

10 132 -1,17 103 0,40 235 -0,48

Total 2223 -4,37 121 -0,26 2344 -4,16

Stratum

Promoted students Retained students Total

=

3. Results

– Within-stratum balance in 97% of the observed pre-retention student, class and school characteristics

→ Retained and promoted children within a certain stratum are equivalent (within sampling fluctuations) in terms of risk factors preceding retention

3. Results

• Same-grade comparisons

– On average:

during their retention year, Grade 1 retainees show a similar psychosocial functioning in comparison to younger grade-mates who are at similar risk of being retained

but … over time they (mostly) grow significantly slower they end up showing more hyperactive behavior, feeling less well at school etc.

! One exception: popularity among classmates

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Ag

gre

ssiv

e b

ehav

ior

Retained

Promoted

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Hyp

erac

tive

beh

avio

r

Retained

Promoted

signsign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Aso

cial

beh

avio

r

Retained

Promoted

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Ind

ep

en

de

nt

pa

rtic

ipa

tio

n

Retained

Promoted

signsign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Sc

ho

ol w

ell-

be

ing

Retained

Promoted

signsign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Se

lf-c

on

fid

en

ce

Retained

Promotedsign

3. Results

! One exception !

1

2

3

4

5

6

0 1 2 3 4 5

Time

Po

pu

lari

ty a

mo

ng

cla

ss

ma

tes

Retained

Promoted

signsign

3. Results

• Same-age comparisons

– On average:

Grade 1 repeaters would have developed a similar or even better psychosocial functioning, had they been promoted to Grade 2 instead, both in the short and long run

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Po

pu

lari

ty a

mo

ng

cla

ss

ma

tes

Retained

Promoted

signsign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Ag

gre

ssiv

e b

ehav

ior

Retained

Promoted

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Hyp

erac

tive

beh

avio

r

Retained

Promoted

sign

sign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Aso

cial

beh

avio

r

Retained

Promoted

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Ind

ep

en

de

nt

pa

rtic

ipa

tio

n

Retained

Promotedsignsign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Sc

ho

ol w

ell-

be

ing

Retained

Promoted

sign

3. Results

1

2

3

4

5

6

0 1 2 3 4 5

Time

Se

lf-c

on

fid

en

ce

Retained

Promotedsign

4. Conclusions and discussion

• Overall, Grade 1 retainees do not seem to benefit much from their retention year

– over time they grow slower compared to grade-mates, making them end up feeling less well at school etc. at the end of elementary school

– while they would have developed a similar or even better psychosocial functioning, had they been promoted to Grade 2 instead

→ Practical implication: Our results call the practice of Grade 1 retention in Flanders into question

4. Conclusions and discussion

• Future research is needed

– Sensitivity analyses

– Other outcome: growth in math and reading skills

– Moderating effects: provision of additional support

--------------------------------------------------------------------------------------------------------------------

2ND BIENNIAL MEETING OF THE EARLI SPECIAL INTEREST GROUP 18 

“Educational Effectiveness: Models, Methods and Applications” 

Leuven, Belgium 25-27 August 2010

http://www.sigee2010.org

--------------------------------------------------------------------------------------------------------------------

Keynote lectures by Prof. Dr. Stephen Raudenbush, Prof. Dr. Robert E.Slavin, Prof. Dr. Jürgen Baumert and Prof. Dr. Jan-Eric Gustafsson

   

Thank you for your attention!

Any suggestions or comments are welcome:

[email protected]