台灣教育長期追蹤資料庫
Taiwan Education Panel Survey
(TEPS)
Ping-Yin Kuan
National Chengchi University
11/16/2004
Main Features
A national longitudinal survey project
collecting data from students of junior high
cohort and senior high cohort from 2001 to
2007. It also collects data from parents,
teachers, and school administrators.
To gain a systematic understanding of the
main factors affecting students’ learning.
To provide an important resource for both
academic research and policy formulation.
Project Background
Lack of good data sets to inform
educational reform policies.
Past educational researches tended to
be cross-sectional, limited in sample
sizes, based on adult population, limited
in model specifications, gave
inconsistent findings, and did not
address basic but critical questions.
Needs international comparisons.
Funding Agencies
Ministry of Education
National Science Council
Academia Sinica
Research Team
Principal Co-investigators and
Investigators: research fellows are
recruited from
Ly-Yun Chang various academic
institutions in
and Taiwan. Yes, there
Tony Tam are UWM graduates
(Academia Sinica) in the team.
Project Concerns
Theoretical Concern Policy Concern
Learning Effects, Educational
Behavioral and Opportunity
Psychological
Consequences
of Schooling School Quality
Institutions
Theoretical Framework
Y = f (A, O, E)
Effects
Ability (A) Analytical
Opportunity (O) Ability (Y1)
Effort (E) Behavior (Y2)
Health (Y3)
Research Possibility (1)
Learning Learning Learning
Y1 Effects Effects Effects
Mental Mental Mental
Y3 Health Health Health
Time 1 Time 2 Time 3
Research Possibility (2)
Learning Learning Learning
Y1 Effects Effects Effects
E Effort Effort
Effort
Time 1 Time 2 Time 3
Educational System in Taiwan
Senior University Graduate
High K13-K16 K17+
K10-K12
3-Year
Vocational College
K10-K12 K13-K15
Elementary Junior
K1-K6 High 2-Year
K7-K9 College
K13-K14
5- Year College
K10-K14
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Basic Research Design
Samples:
-- Junior high sample (1st year students)
-- Senior high/vocational high sample (2nd
year students)
-- Junior college (2nd year students)
Multiple perspectives (student, parent, and
teacher) on selected student, parent, teacher,
and school attributes.
Longitudinal study and inter-cohort
comparisons with comparable sampling
designs.
Project Timetable
Nov. 1999 to Dec. 2000 Year
Semester 2001 2002 2003 2004 2005 2006 2007
1st 2nd 1st 2nd 1st 2nd 1st 2nd 1st 2nd 1st 2nd 1st 2nd 1st 2nd 1st
Data Data Data Data Data
collection collection collection collection Data
Data cleaning, Data cleaning, cleaning,
Junior high
cleaning
Junior Junior Senior Senior
sample
compilation and compilation compilation and
high high high high compil-
Fist year follow-up planning Third and follow-up Second and Third
(7th year year year ation
grade) (9th planning (11th follow-up (12th
grade) grade) grade)
planning
Data Data cleaning, Data
collection collection
compilation
Senior high
Senior and follow-up Senior
sample
Preparatory high high
Data cleaning
Second
planning Third and
operations year year compilation
(11th (12th
grade) grade)
Data Data cleaning, Data
collection collection
Junior college
compilation
Junior and follow-up Junior
sample
Data cleaning
college planning college and
Second Third
year year compilation
Sampling Design
Concerns Design
- Stratified by urban/rural,
- Causal Analysis
public/private, and
- Multi-level Analysis school types
- Attrition in Follow- - Sample school
programs first, then
ups classes, and then
students. In principle,
4 classes and 15
students in each
class were sampled.
- Oversample certain
populations
Sample Size (2001/2003)
Program Types
Total Junior Senior Junior
Senior High
High Vocational College
Total number of Regular Comprehensive
schools/programs 546 338 62 27
sampled 163 49
Actual number of
schools/programs 539/ 333/ 159/ 48/ 62/ 26/
for which data 535 333 158 46 64 26
were available
Actual number of
2,303/ 1,244/ 573/ 130/ 260/ 96/
classes for which
3,213 1,930 733 156 283 111
data were available
Actual number of
39,336/ 20,004/ 8,719/ 2,062/ 4,066/ 4,485/
students for whom
37,133 18,903 8,264 1,947 3,842 4,177
data were available
Actual number of
6,601/ 3,580/ 2,728/ 312/
teachers for whom
6,558 3,845 2,486 245
data were available
Data Collection
Ability tests: For students only.
Questionnaires: Completed by students,
parents, teachers and school
administrators.
Questionnaire Design
Students Parents Teachers Schools
I. Daily Schedule at I. Family Characteristics I. Questions to be I. Questions to be
School Answered by All Answered by the
II. About the Student Teachers Principal
II. Home Environment
III. Student’s Life During II. Questions to be II. Questions to be
III. School Environment the Primary School Answered by the Answered by the
and Junior High Homeroom Teacher Director of Academic
IV. Extracurricular School Years Affairs
Activities and Peers III. Questions to be
IV. Relationship Between Answered by Class III. Student Affairs
V. Personal the Family and the Teachers (Chinese,
Backgrounds, Current School English, and Math) IV. School Funding and
Educational Equipment
Aspiration, V. Expectations of the VI. Civic Orientation
Self-Evaluation Student V. Personnel Matters
VI. Civic Orientation VI. Civic Orientation
Ability Tests
Measurement of overall analytical ability
or problem solving ability that would
reflect a student’s learning achievement
and growth.
Tests emphasize the ability to solve
problems through analysis and deduction
rather than through rote learning.
Test modules include general deductive
reasoning, science, mathematics, and
languages.
Test results are estimated ability scores
based on Item Response Theory.
Some Preliminary Research
Findings
Two basic issues:
What is the role of the family in
the making of educational
inequality?
What is the relationship between
academic achievement and
adolescents’ mental health?
The First Issue:
Empirical Questions (1): Effects of
Parental SES?
How does parental socioeconomic status,
measured in terms of income and
education, matter for the cognitive
achievement of students? (How does financial
constraint compare to parental education?)
How do parental SES effects vary across
grades?
How different are parental SES effects in
Taiwan and the U.S.?
Measurement Strategies
Constraint:
Different countries have different classifications
to begin with.
Compromise:
To facilitate cross-national comparisons,
adopt the same number of categorical income
and parental education for Taiwan and the U.S.
Categorical measures of family background to
allow for potentially nonlinear effects.
Data
TEPS NELS:88
-- 2001 (Fall) -- 1988 (Base year)
8th Graders
7th Graders (Junior
High Cohort)
-- 1990 (1st Follow-Up)
11th Graders (Senior 10th Graders
High Cohort)
-- 2003 (Spring) -- 1992 (2nd Follow-Up)
12th Graders (Senior 12th Graders
High Cohort, 1st
Follow-Up)
CROSS-SECTION
ANALYSIS: IRT Score
Figure 1A. Family Income, Parental Education
& Student's Achievement: K7
(TEPS 00, N=12503) IRT score
Income/Education Income+Education
120 116
98
100 95
91 91
86
81
80 76
63
59
60
46 46
40
40 34 32
28 25
18
20
0
=
le
HS
Co
Un
G
5-
ss
ra
5
1
k
-
-
20
lle
iv
k
15
20
0k
du
th
k
ge
k
k
at
an
e
HS
Figure 2A. Family Income, Parental Education
& Student's Achievement: K11
(TEPS 00, N=11445) IRT score
Income/Education Income+Education
117
120
100 94
87 90
86 87
80
67
55 57
60 52 53 53
37 39
40
23 26
17 18
20
0
=
le
HS
Co
Un
G
5-
ss
ra
5
1
k
-
-
20
lle
iv
k
15
20
0k
du
th
k
ge
k
k
at
an
e
HS
Figure 3A. Family Income, Parental Education
& Student's Achievement: K12
(TEPS 00, N=11447) IRT score
Income/Education Income+Education
120
102
100
78 80
80 73 75
69
58
60
45 44 46
40 43
40 31
29
23
19 17
20 13
0
=
le
H
C
U
G
5-
S
ol
ni
ss
ra
5
1
-
-
k
20
k
15
20
le
v
0k
du
th
ge
k
k
k
at
an
e
H
S
Figure 3C. Net Family Income Effects: K7, K11, & K12 in TEPS
％
R-sq
K11 vs K7 = 0.95
K7 K11 K12 K11 vs K12 = 0.97
400 K12 vs K7 = 0.89
300
200 178
166 161
159
135
125 124 119
113
93 87
100 85
38 33 35
0
=200k
Figure 3D. Net Parental Education Effects: K7, K11, & K12 in TEPS
％
R-sq
K11 vs K7 = 0.92
443
K11 vs K12 = 0.998
425
K7 K11 K12 K12 vs K7 = 0.91
400
339
329
300
232
215
200
178
104
100 79
59
40 41
0
less than HS HS J. College Univ Graduate
Figure 1B. Family Income, Parental Education
& Student's Achievement: K8
(NELS 88, N=21930) IRT score
Income/Education Income+Education
145
140 131
120 111 107
102
100
84
79
80 72
62 60 62
56
60
45 42
40
40 30
23 26
20
0
=
le
H
C
U
G
S
ol
ni
ss
ra
0k
-
-
-
-
75
20
35
50
75
le
v.
du
th
k
ge
k
k
k
k
at
an
e
H
S
Figure 2B. Family Income, Parental Education
& Student's Achievement: K10
(NELS 88, N=14801) IRT score
Income/Education Income+Education
159 157
160
140
125
120 113 113
98 96
100 88
80 71 69 69 68
60 54
43 45
37 38
40
21
20
0
=
le
H
C
U
G
S
ol
ni
ss
ra
0k
-
-
-
-
75
20
35
50
75
le
v.
du
th
k
ge
k
k
k
k
at
an
e
H
S
Figure 3B. Family Income, Parental Education
& Student's Achievement: K12
(NELS 88, N=12321) IRT score
Income/Education Income+Education
166
160
160
140 130
125
120 113
97 96
100 89
80 71 72
63 60
60 52 50
38 40
40 31
24
20
0
=
le
H
C
U
G
S
ol
ni
ss
ra
0k
-
-
-
-
75
20
35
50
75
le
v.
du
th
k
ge
k
k
k
k
at
an
e
H
S
Summary of Findings
What appears to be strong family income
effect in Taiwan, despite the emphasis by
recent critics of Taiwanese education, is
largely spurious (of parental education).
Parental education effects are remarkably
stable across high school grades.
The qualitative and quantitative results are
surprisingly similar across two strikingly
dissimilar societies.
The First Issue:
Empirical Questions (2): Effects of
Family Structure?
How do types of family structure
affect the cognitive achievement of
students?
-- Strong evidence has emerged that single-
parent and stepparent families have adverse
effects on children’s educational achievement.
-- Some studies in the U.S. also found that
children of single-parent families with
cohabitating grandparent(s) performed quite
similarly to those of intact families.
The First Issue:
Empirical Questions (2)
How about co-residing grandparent(s)
in an intact family? Will they bring
similar positive educational
advantage to their grandchildren?
-- In Taiwan, not only nuclear intact families are
still the dominant family type, but the
multigenerational intact families composed by
two biological parents, unmarried children, and
at least a grandparent still consist about 11%
of households in Taiwan (2000 census).
Why Does Family Structure Matter to
Children’s Achievement?
Economic resources: Non-intact families are
often trapped in poverty or have greater
economic burden.
Socialization resources: Non-intact families
are less able or less likely to provide a good
environment for children in terms of
educational involvement and educational
aspiration.
Network resources: Non-intact families have
fewer network ties for obtaining information
and other types of support related to
children’s learning.
What might a grandparent bring to the
family? The case in Taiwan
Economic resources?
Even though the rate of cohabitating with
older parents is declining, non-cohabiting
adult children still feels obligated to support
their parents financially.
Socialization resources?
Co-residing grandparents may provide more
psychological support for the grandchildren,
convey parents’ expectation, give advice to
the grandchildren, and constantly monitor the
grandchildren’s activities at home.
What might a grandparent bring to the
family? The case in Taiwan (cont’d)
Network resources?
The presence of grandparents may give
additional linkage to relatives,
communities, and schools and, hence,
contribute to grandchildren’s learning.
Data and Method
DATA: Two cohorts of TEPS– 7th Graders (N =
12,442) and 11th Graders (N = 12,320)
Measures
-- Dependent Variables: IRT Ability Score
-- Independent Variables:
5 types of family structure –
(1) Nuclear intact (reference) (65%; 67%)
(2) Multigenerational intact (17%; 18% )
(3) Multigenerational single-parent (3 %; 2%)
(4) Single-parent (8% )
(5) All other types of non-intact (15%; 13%)
Data and Method (cont’d)
-- Indep. Var.:
Indicators of 3 types of resources:
Economic: Monthly family income
Socialization: Attend school events; talk about
school; talk about inner thoughts; checking
homework; educational expectation
Network: visit relatives; know other parents
Control variables: Sex, sib size; ethnicity; parents’
educational level
Data and Method (cont’d)
Method: OLS regression
Model 1: Types of family structure (gross
effects)
Model 2: Model 1 + control variables
Model 3: Model 2 + monthly family income
Model 4: Model 3 + indicators of socialization
resources
Model 5: Model 4 + indicators of network
resources
R2 of 5 Regression Models for Two Cohorts
R2 The 7th grader
0.35
The 11th grader
0.3
0.25
0.2
0.15
0.1
0.05
0
model-1 model-2 model-3 model-4 model-5
Regression Models
Effects of Types of Family Structure in 5
Regression Models
multigenerational intact multig eneratio nal in tact
multigenerational with single-parent multig eneratio nal with sin gle-parent
b b sin gle parent
single parent
0.2 0.2 all oth er ty pes
all other types
0.1 0.1
0.0 0.0
-0.1 -0.1
-0.2 -0.2
-0.3 -0.3
-0.4 -0.4
-0.5 -0.5
-0.6
-0.6
-0.7
-0.7
-0.8
-0.8
model-1 model-2 model-3 model-4 model-5
model-1 model-2 model-3 model-4 model-5
7th Graders 11th Graders
Summary of Findings
The impact of family structure is mediated by 3
types of resources. The mediating variables
related to economic resources and parental
involvement, however, have larger effects for
the older cohorts than for the younger cohorts.
The addition of a grandparent is beneficial to
children’s educational achievement. This
positive contribution, however, depends on the
type of family structure.
The effects of family structure types are all
smaller for the older cohort.
The Second Issue:
Empirical Questions
Would high academic achievement and
expectation induce poorer mental health?
-- Previous research found a weak positive
relationship.
Does high family SES induce poorer mental
health?
-- High SES parents tend to have higher academic
expectation and be more involved in the children’s
education, which in turn make their children perform
better academically.
-- Previous studies have found positive relationship
between SES and mental health. But some studies
also found high SES or high achieving students have
more distress. This relationship may be due to the
higher achievement pressure of the high family SES.
Empirical Questions (cont’d)
How do different parenting styles
and parental involvement strategies
affect adolescent’s mental health
then?
-- Authoritative parenting style (responsive but
firm control) has been found to be positive to
children’s academic achievement and adjustment
in general.
-- Psychological control, on the other hand, has
been found to be related to adolescents’ poor
psychological and behavioral outcomes. But no
report is on its effect on academic achievement.
-- Specific parental involvement strategies in
education include all three parenting dimensions:
support, behavioral control, and psychological
control (such as high parental expectation).
Parent's Edu. Expectation and Parental Edu. Level
Junior Col
80
Tech.
Parent's Edu. Expectation (%)
73.3
Col/Univ
70 Reg.
63.1 Col/Univ
60 Grad. Sch
56.3
48.4 49.7
50 47.6
45.3
42.2
39.3
40
32.2
30 28.1
27.2
26 24.9 25.2
20.5 21.6
19.6 18.1 18.3
20 17.8 17.2
14.4
14.2 15.7
13.6
10 8.1 7.1 5.5 5.9
1.3 1.9 1.2 2 0.7 0
0
1 1.5 2 2.5 3 3.5 4 4.5 5
Parental Edu. Level
Data and Method
Data: 11th graders (N = 11,515)
Measures
-- Dependent Variable: 14 items selected
from The Symptom Checklist-90-Revised
(SCL-90-R) that measured frequencies of
depressive, anxious, aggressive, and
psychosomatic symptoms and suicidal
ideation. A factor score was derived from
the 14 items by using the confirmatory
factor analysis modeling.
Data and Method (cont’d)
-- Independent Variables:
1. Academic achievement: IRT ability score.
2. Family SES: Parents’ educational level and
monthly income.
2. Authoritative parenting: items related to parents’
acceptance, non-punitive behavior, and listening
to inner thoughts.
3. Parental involvement in education:
a. Involvement related to support/warmth and
behavioral control including ‘helping with
school work’, ‘checking school work’, and
‘supervision after school’.
b. Involvement related to psychological control
including ‘talks about future schooling plans’
and ‘talks focused on academics’.
Data and Method (cont’d)
-- Control Variables:
1. Student’s sex
2. Stressful family events experienced: Parents’
divorce, separation, or death; parents very ill;
parents with psychological illness; alcoholic
parents; sudden economic fall of the family.
Method: OLS regression
Model 1: Control + Family SES
Model 2: Control + SES + IRT score + (IRT score) 2
Model 3: Control + Parenting behavior and
educational involvement
Model 4: Full model
Figure 1. Prediction of Mental Health with Controls Only
40.00
20.00
mental health
0.00
-20.00
-40.00
-3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 4.00
Academic Achievement
Figure 2. Prediction of Mental Health Adjusted for SES Factors
40.00
30.00
20.00
Mental Health
10.00
0.00
-10.00
-20.00
-30.00
-3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 4.00
Academic Achievement
Summary of Findings
Higher family SES has negative effects
on mental health. But the effect mostly
vanishes when academic achievement
is controlled.
Academic achievement has negative
effects on mental health. Although the
quadratic term is significant, the trend is
linear most of the time.
Summary of Findings (cont’d)
Parental support and behavioral control
generally have positive effects on both
mental health and academic achievement.
Parental acceptance (dimension of warmth
and support) has effects positive on mental
health and negative on academic
achievement. The latter effect falls into
insignificance once SES is controlled. Non-
punitive parenting, however, has positive
effects on either achievement or mental
health.
Summary of Findings (cont’d)
Parents talking about schooling and
occupation plans (psychological control)
is positively related to academic
achievement, but negatively related to
mental health.
The effect sizes of parenting behaviors
remained very much the same after
academic achievement is controlled.
Access to TEPS: A Public Asset
Data of the 1st wave (2001) has been released for
public access (http://www.teps.sinica.edu.tw). The
first follow-up data (2003) will be released soon.
Three levels of access:
-- Public access: Online application; no school and
class id; 70% of the original sample.
-- Restricted access: Restricted to academic and
governmental institutions; needs to sign an
agreement of confidentiality; could study class effects,
but not school effects.
-- On-site access: Further restriction (at least a Ph.D.
candidates with advisors’ endorsement). Nearly full
access to the data.
Access to TEPS (cont’d)
English translation of the questionnaires
and various handbooks is underway.