Main Topic: How are social, economic, ethnic and gender inequalities generated and maintained across

1
Center for Research on Inequalities and the Life
Course (CIQLE)

• Main Topic How are social, economic, ethnic and
  gender inequalities generated and maintained
  across the life course?
• Methods Eclectic and catholic, but major focus
  on quantitative, longitudinal analysis of
  individual and household microdata

2
CIQLE

• Agenda Support for research
• Data archive for the German Life History Study
  and comparable micro data from the U.S. and other
  countries
• Assistance with using these and other data that
  fall within the expertise of CIQLE affiliates
• Methodological workshops
• Outreach to members of the Yale community
  interested in research on social inequality

3
Events to come

• October 27th Richard Eibach (Psychology) on his
  work on social judgment
• December 1st Kim Blankenship (CIRA) on her work
  on social inequality and AIDS/HIV
• Fall workshop Methodological workshop on Event
  History Analysis, with Uli Mayer (November)
• Spring workshop Models for Panel Data (unless
  you have other needs/suggestions)

4
CIQLE people

• Karl Ulrich Mayer (on leave) Life course and
  aging, social mobility, sociology of education,
  welfare state and social policy
• Averil Clarke Family formation, race, class, and
  gender stratification
• Hannah Brückner Gender inequality, labor
  markets, social policy, social networks,
  adolescent behavior

5
Data Analysis with the National Longitudinal
Study of Adolescent Health (Add Health)

• Hannah Brückner
• Yale University
• Center for Research on Inequalities and the Life
  Course

6
What you can expect from this talk

• Add Health features
• Design
• Special samples
• Network data
• Some examples of research using Add Health
• Some limitations and problems

7
Some Sources Used in this Presentation

• Add Health webpage
• http//www.cpc.unc.edu/projects/addhealth/
• James Moodys webpage
• www.sociology.ohio-state.edu/jwm
• Peter Bearmans webpage
• www.iserp.columbia.edu/people

8
Sample design

• Stage 1
• School sample stratified, random sample of all
  high schools in the United States. A school was
  eligible for the sample if it (1) included an
  11th grade and (2) had a minimum enrollment of 30
  students. A feeder school a school which sent
  graduates to the high school and which included a
  7th grade was also recruited from the
  community. High schools were stratified into 80
  clusters by
• Region--Northeast, Midwest, South, West
• Urbanicity--urban, suburban, rural
• School size--125 or fewer, 126-350, 351-775, 776
  or more students
• School type--public, private, parochial
• Percent white--0, 1 to 66, 67 to 93, 94 to 100
• Percent black--0, 1 to 6, 7 to 33, 34 to 100
• Grade span--K to 12, 7 to 12, 9 to 12, 10 to 12
• Curriculum vocational/technical, alternative,
  special education

9
Participating High Schools

10
Add Health Instruments and Other Study Elements

• 90,118 In-school Questionnaires (September
  1994-April 1995)
• 164 School Administrator Questionnaires--Wave I
  (September 1994-April 1995)
• 20,745 adolescent In-home Interviews--Wave I
  (April 1995-December 1995)
• Add Health Picture Vocabulary Test scores (April
  1995-December 1995)
• 17,700 Parent Questionnaires (April 1995-December
  1995)
• 125 School Administrator Questionnaires--Wave II
  (May 1996-June 1996)
• 14,738 adolescent In-home Interviews--Wave II
  (April 1996-August 1996)
• Community contextual dataset
• In-school friendship network dataset
• Geographic location of households around a
  central point in the community
• 15,172 young adults in wave III (2000-2001)
• Transcript data collected from last school
  attended, to be released next year

11
virginity pledges
12
In-school component

• 145 middle, junior high, and high schools
  participated
• 90,118 students in grades 7 to 12 completed a 45
  minutes self-administered questionnaire
• Data were collected from September 1994 through
  April 1995
• Each participating school was asked to complete a
  school administrator questionnaire

13
In-school component, cont.

• Parental consent was required to list student
  names in a directory and to allow students to
  participate in the study.
• There was no "make-up" day for students not
  present on the day of administration. Parents
  were informed in advance when the questionnaire
  administration would occur and could direct that
  their children not participate.
• Unless otherwise directed by the school, passive
  consent forms were used (it was assumed that a
  parent granted permission unless the form was
  returned with a signature that indicated
  otherwise).
• Some schools required active consent forms (the
  form had to be returned with a signature
  indicating that permission was granted).

14
In-school component, cont.

• The questionnaire included questions on
• Students and parents background
• Relationship to parents
• Academics and extra-curricular activities
• General health status and health-related
  behaviors
• Rudimentary Scales (self-esteem, depression)
• Delinquency, risk behavior, aspirations and
  expectations

15
In-school component, cont.

• Information on up to five female and male friends
• Friends could be selected from school roster
• Information about nature and frequency of contact
  with the nominated friends
• later more on this

16
School Administrator Survey

• General characteristics of school and student
  body
• Teacher training and turnover
• Curriculum, school services, and programs
• Disciplinary policies

17
In-home Survey

• Stage 2
• An in-home sample of 27,000 adolescents was drawn
  consisting of a core sample from each community
  plus selected special over-samples. Adolescents
  were drawn from a sampling frame which included
  students who were on the school roster and
  students who participated in the in-school
  component.
• Eligibility for over-samples was determined by
  an adolescent's responses on the In-school
  Questionnaire. Adolescents could qualify for more
  than one sample.

18
In-home Survey, cont.

• Written informed consent was obtained from
  adolescent and parent
• Computer-assisted personal interviews
• Sections with sensitive questions were asked in
  self-administered Audio CASI
• Data collection lasted from April 1995-December
  1995
• Core Sample 12,105 adolescents representative of
  adolescents in grades 7 to 12 during the
  1994-1995 school year in the United States

19
In-home Survey, cont.

• Special Samples
• Saturated schools 2,526 adolescents (in addition
  to the 200 students selected for the core) from
  schools in which all students were selected for
  in-home sample
• Disabled sample 471 adolescents who reported
  having a limb disability
• High SES African-American sample 1,038 black
  adolescents with a parent with a college degree
• Chinese 334 adolescents
• Cuban 450 adolescents
• Puerto Rican 437 adolescents

20
In-home Survey, cont.

• Genetic Sample Adolescents pairs with varying
  degrees of relatedness (except for twins, all had
  to be enrolled in grades 7 to 12 at the time of
  the sample selection)
• Twins 1,981 adolescents (incl. a couple of
  triplets)
• Full sibling 1,186 adolescents
• Half sibling 783 adolescents
• Non-related adolescent 415 adolescents
• Sibling of twins 162 adolescents

21
In-home Survey, cont.

• Diet, physical activity, height/weight
• Morbidity, injury, chronic and disabling
  conditions, health care utilization
• Pubertal development
• Sexual behavior, contraception, pregnancy
• Contraceptive knowledge
• Dating behavior, romantic and sexual
  relationships

22
(No Transcript)
23
In-home Survey, cont.

• Peer networks
• Decision-making processes
• Family composition and dynamics
• Academics, school integration
• Educational aspirations and expectations
• Self efficacy, self-esteem, feelings scale,
  personality, relationship to parents siblings

24
In-home Survey, cont.

• Fighting and violence, victimization, substance
  abuse
• Suicidal ideation and attempts
• Neighborhood
• Employment
• ..

25
In-home Survey, cont.

• Add Health Picture Vocabulary Test (abridged
  version of Peabody, age standardized)
• Spatial data GPS generated longitude and
  altitude of household, used to calculate distance
  from central point in community and to link
  contextual data

26
In-home Survey, cont.

• Parental interview (in most cases, with mother)
• Mothers marital history, employment, and
  education
• Household income, health insurance status
• Neighborhood characteristics (social capital)
• Parent-adolescent communication and interaction
• Health behavior of mother, spouse, and adolesent
• Educational aspirations and expectations for the
  adolescent

27
In-home Survey, cont.

• Contextual data base (from census and other
  data), including characteristics of census
  tract/block, county, and/or state
• Socio-economic variables, income and poverty
• Crime rates, tobacco use data, STD rates
• Social programs and policies
• Demographic and household characteristics
• ..

28
(No Transcript)
29
In-home Survey, Wave II

• 14,738 adolescent In-home Interviews (April
  1996-August 1996)
• Eligible were all wave 1 participants except
  seniors
• One-year observation window for prospective
  analysis
• With some exceptions, wave 1 questionnaire was
  administered again

30
(No Transcript)
31
Wave III

• Wave III data collection, conducted in 2001 and
  2002, includes in-home interviews with original
  respondents (now young adults) and in-home
  interviews with their partners. Data is available
  on 15,172 respondents. Wave III included these
  specific aims
• locating 1995 Wave I Add Health in-home
  respondents
• collecting longitudinal data on Add Health
  respondents
• collecting data on subsamples of married,
  cohabiting, and dating partners of respondents
• collecting specimens of saliva and urine for
  assays of HIV and STDs, in order to develop
  prevalence estimates
• collecting geocodes for respondents' addresses at
  the time of interview

32
Wave III, cont.

• Add Health Picture Vocabulary Test
• Overview and Demographics
• Household Roster and Residence History
• Parental Support and Relationships
• Marriage/Cohabitation History and Attitudes
• Economics and Personal Future

33
(No Transcript)
34
Wave III, cont.

• Retrospective ADHD
• Relationships with Siblings
• Friends
• Education, including transition to college and
  other school, detailed information on majors
• Labor Market Experience and Active-Duty Military
  Service

35
Wave III, cont.

• General Health and Diet
• Access to Health Services, Health Insurance
• Illnesses, Medications, Physical Disabilities
• Social Psychology and Mental Health
• Delinquency and Violence
• Involvement with the Criminal Justice System
• Tobacco, Alcohol, Drugs, Self-Image
• Mistreatment by Adults

36
Wave III, cont.

• Sexual Experiences and STDs
• Relationships, completed and current pregnancies,
  life births
• Children and Parenting
• BEM Inventory
• Propensity for Risk

37
Wave III, cont.

• Civic Participation and Citizenship
• Religion and Spirituality
• Gambling
• Daily Activities
• Biological Specimen Participation
• Interviewer's Report

38
(No Transcript)
39
Social Networks

• Social network analysis is
• a set of relational methods for systematically
  understanding and identifying connections among
  actors
• a body of theory relating to types of observable
  social spaces and their relation to individual
  and group behavior.

Source Jim Moody, Introduction to Network
Analysis
40
Types of Relational Data in Add Health

• Friendship nominations from in-school survey (up
  to 5 male and 5 female friends)
• Friendship nominations from in-home surveys (1
  male and 1 female friend, except for saturated
  schools)
• Romantic and sexual partner nominations (up to
  six partners in wave 1 and 2, in-home surveys)
• Shared participation in extra-curricular
  activities (in-school survey)
• Transcript data shared enrolment in a specific
  class level (but not sharing the same class room)

41
Types of Questions for which we need network data
analysis

• 1. Networks as dependent variables
• Explore network structure to understand a process
  of diffusion
• Discover social processes that account for the
  observed network structures
• Model the presence or absence of specific ties in
  a network as a function of network structures and
  social processes

42
Example 1

• AMERICAN JOURNAL OF SOCIOLOGY, JULY 2004
• Chains of Affection
• The Structure of Adolescent Romantic and Sexual
  Networks
•  
•  Peter S. Bearman, Columbia University
•  James Moody, Ohio State University
•  Katherine Stovel, University of Washington
• Data for this paper are drawn from the National
  Longitudinal Study of Adolescent Health (Add
  Health), a program project designed by J. Richard
  Udry and Peter Bearman, and funded by a grant
  HD31921 from the National Institute of Child
  Health and Human Development to the Carolina
  Population Center, University of North Carolina
  at Chapel Hill, with cooperative funding
  participation by the following agencies The
  National Cancer Institute The National Institute
  of Alcohol Abuse and Alcoholism the National
  Institute on Deafness and other Communication
  Disorders the National Institute on Drug Abuse
  the National Institute of General Medical
  Sciences the National Institute of Mental
  health the Office of AIDS Research, NIH the
  Office of Director, NIH The National Center for
  Health Statistics, Centers for Disease Control
  and Prevention, HHS Office of Minority Health,
  Centers for Disease Control and prevention, HHS,
  Office of the Assistant Secretary for Planning
  and Evaluation, HHS and the National Science
  Foundation. The authors thank Douglas White,
  Martina Morris, Mark Hancock, and J. Richard Udry
  for helpful comments on previous drafts of this
  paper. Corresponding author. Institute for
  Social and Economic Research and Policy. 814 SIPA
  Building, Columbia University, New York NY 10027.
  Email psb17_at_columbia.edu

43
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
44
Contact structure
Local vision
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
45
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
Global Vision A sexual and romantic network for
a single high school
63
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
46
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
Simulated and Observed Network Features
Simulated networks preserve observed degree
distribution
47
Hypothetical Romantic Cycle
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
48
Indirect Pathways
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
49
Indirect Cores Removing cut-point ties in
indirect relationship graph
Shaded nodes are members of indirect cores
Bearman, Moody, Stovel Chains of Affection,
AJS 2004
50
What Are the Implications of This Work, for
Intervention?

• Very small changes in behavior anywhere in the
  system can have very significant impacts on the
  structure of disease diffusion.
• But, we cannot tell ex ante where such cuts will
  have the biggest impact. One predictor that would
  seem important is not at all important number
  of sex partners

Bearman Adolescent Sexual Behavior and STD
Acquisition Some Recent Research Findings (CDC
Conference)
51
What Are the Implications of This Work, for
Intervention?

• 3. Consequently, intervention to stem adolescent
  STD diffusion, in contrast to adults, ought to be
  broad-based. Targeted interventions may change
  behavior, but not where it counts.
• 4. Adolescents are different than adults.
  Importing adult models of sexual networks into
  the adolescent context is misleading.

Bearman Adolescent Sexual Behavior and STD
Acquisition Some Recent Research Findings (CDC
Conference)
52
Types of Questions for which Network Analysis
Might Be Useful

• 2. Networks as independent variables
• Compare the network structures arising from
  different social settings to explain outcomes at
  the micro or meso level
• Investigate the impact of a persons position in
  the network on his or her behavior or well-being
• Investigate the impact of various relational
  contents in a persons network on his or her
  behavior or well-being
• Compare sources of social influence across
  various levels of social context

53
Measuring Network Context Patterns Positions

• Pattern measures capture some feature of the
  distribution of relations across nodes in the
  network. These include
• Density of all possible ties actually made
• Reciprocity likelihood that given a tie from i
  to j there will also be a tie from j to i.
• Transitivity extent to which friends of friends
  are also friends
• Hierarchy Is there a status order to
  nominations? How is it patterned?
• Clustering Are there significant groups? How
  so?
• Segregation Do attributes (such as race) and
  nominations correspond?
• Distance How many steps separate the average
  pair of persons in the school? Is this larger or
  smaller than expected?
• Block models What is the implied role structure
  underlying patterns of relations?
• These features (usually) require having
  nomination data from each person in the network.

Moody Introduction to Network Analysis
54
Jim Moody Introduction to Network Analysis Peter
Bearman Social Structure of Suicide
Example 2 Nets as Independent Variable
Relational Structures and Forms of Suicide
Regulation
Low
High
High
Anomic
Altruistic
Integration
Low
Egoistic
Fatalistic
55
Measuring Isolation and Anomie.
Source Jim Moody Introduction to Network
Analysis
56

• Net of a host of individual level
  characteristics, including depression, the
  content and structure of adolescent networks
  matter, but more strongly for girls
• Girls who are socially isolated or occupy an
  anomic position are more likely to ideate suicide
• All adolescents are more likely to report
  suicidal ideation if they have friends who
  attempted suicide

Source Bearman and Moody 2004 Suicide and
Friendship Among American Adolescents, American
Journal of Public Health Vol 94, No. 1
57
Measuring Network Context Composition

• Composition measures capture characteristics of
  the population of people within a given network
  level. These include
• Heterogeneity How dispersed are actors with
  respect to a given attribute?
• Means What is the mean GPA of egos friends? How
  likely is it that most of egos friends will go
  to college?
• Dispersion What is the age-range of people ego
  hangs out with?
• These features can often be measured from the
  simple ego network.

Moody Introduction to Network Analysis
58
Example 4 Levels of Peer Influence Peter
Bearman and Hannah Brückner Power in Numbers
(1999) Research commissioned by The Campaign to
Prevent Teen Pregnancy Investigates the
influence of peers across different levels of
social relations on girls sexual debut and
pregnancy risk
59
Send- and receive network Having high risk male
friends doubles the risk of getting pregnant
between wave I and wave II compared to those with
low-risk male friends. Not having friends is not
protective (other things equal).
Bearman and Brückner 1999
60
Characteristics of the peer network can be as
important as individual characteristics, as shown
here for closeness to parents. In this study,
peer influence was largely additive adding peer
characteristics to the baseline model did
generally not change the coefficients for
individual and family characteristics.
Bearman and Brückner 1999
61
Much peer influence is positive. Low-risk friends
are better than no friends, and better than
high-risk friends. However, no significant
difference between average and high/low risk
friends was found. Before Add Health was
available, many studies focused on best friends
for lack of more extensive data. Often, only
proxy reports about friends were available, but
we know that adolescents tend to overestimate
similarity in behavior between their friends and
themselves.
Bearman and Brückner 1999
62
Positive influences from low-risk friends are
stronger at the level of ego networks compared to
looking only at best friends.
Bearman and Brückner 1999
63
Peer influences are even stronger at the peer
group level. This is a level that adolescents and
their parents do not necessarily see and know
about, because it includes the friends of friends
and their friends. Peer groups were identified
with Jim Moodys CROWD algorithm (more later)
Bearman and Brückner 1999
64
Some Conclusions

• Peer influence operates at multiple levels of
  peer context.
• The immediate circle of friends and larger peer
  groups are more important than are best friends.
• Low-risk female friends protect against negative
  health outcomes, early transition to first
  intercourse and heightened pregnancy risk.
• High-risk male friends heighten risk for early
  sexual debut and pregnancy.
• Peer characteristics are more important
  predictors of pregnancy risk than individual age
  or risk status.
• No effect of characteristics of the leading
  crowd in a school.
• Strong effects of the proportion sexually active
  in the school, even when controlling for
  lower-level peer contexts.

65
Some Limitations and Problem (of many possible)

• Global network data is available only from the
  in-school survey for most schools adolescents
• Availability of in-school data maybe selective
  for some outcomes and groups
• Extent to which true friendship network is
  captured in in-school data varies
• Extent to which it is possible to differentiate
  between social selection and influence varies and
  depends, among other things, on how well we can
  measure important risk and protective factors

66
Network Data Collected in Add Health
In -School Network Data
Example 1. Ego is a matchable person in the
School
Out
Un
Out
Out
Un
Un
M
Ego
M
Ego

M
M
M
M
M
M
True Network
Observed Network
Moody Introduction to Network Analysis
67
Network Data Collected in Add Health
In -School Network Data
Example 2. Ego is not on the school roster
M
M
M
Un
M

Un
M
M
M
M
M
M
Un
Un
Un
True Network
Observed Network
Moody Introduction to Network Analysis
68
Details on some problems (Examples)

• No in-school data for almost 5,400 wave I
  respondents
• Age at sexual debut for some groups is severely
  overestimated when restricting sample to those
  with in-school data
• Some schools had large number of students who
  were not in the roster, and therefore could not
  be nominated as friends in in-school survey
• Some people have more out-of-school friends than
  others, and therefore we cannot accurately
  measure peer context for them

69
(No Transcript)
70
a Includes all respondents with valid in-school
surveys. Includes those who did not nominate
anybody. b Calculated as number of in-school
respondents over total number of nominations.
Does not take item non-response into account,
only survey non-response and nominations of
friends which were not on the roster/wrong AID.
c When excluding schools where less than 70 or
80 of in-school-nominations were going to
friends with AID, the number increases very
slightly to 57-58.
71
(No Transcript)
72
(No Transcript)
73
Add Health specific resources The network data
set

• INDIVIDUAL-LEVEL MEASURES
• Basic Network Descriptors
• Ego-centered Network Measures
• Sociometric characteristics of ego-networks
• Ego-network heterogeneity measures for grade,
  race, and age
• Ego-network behavior/attribute means for
  in-school questionnaire items
• SCHOOL-LEVEL MEASURES
• Measures of Global Network Structure
• Measures of Segregation and Group Salience for
  Grade, Race, and Sex