Longitudinal Data Across Multiple Stages Of Life

1
Longitudinal Data Across Multiple Stages Of Life

• Substantive Applications
• Hanno Petras, Ph.D.
• Johns Hopkins University

2
Collaborators

• Hendricks Brown, USF
• Howard Chilcoat, JHU
• Nick Ialongo, JHU
• Shep Kellam, AIR JHU
• Phil Leaf, JHU
• Bengt Muthen, UCLA
• Prevention Science Methodology Workgroup

3
Paper 1

• Developmental Antecedents and Malleability of
  Antisocial Personality Disorder - Long-term
  Effects of a Universal Classroom Based Preventive
  Intervention

4
Developmental Relationship between Aggressive
Behavior and Antisocial Personality Disorder
(ASPD) in One Cohort of Control Boys (N138)
BIC2516.86 Entropy0.861 of Iterations93
Elementary School
Middle School
Prevalence of ASPD
69.7
35.0
FFall SSpring
10.7
5
Methodological Topics in Growth Modeling

• Impact of particular time points (e.g., in middle
  school) for the prediction of distal outcome,
  over and above class membership (u on Ys)
• Individual intervention impact over and above
  class membership
• Atypical individual development --Outliers
• Modeling of Subgroup variation within Growth
  Modeling (e.g., in control conditions, cohorts)

6
Paper 2

• Specificity/Sensitivity of Predicting
  Developments in Aggression Leading to Violent
  Juvenile Arrest

7
Developmental Relationship between Aggressive
Behavior and Violent Juvenile Arrest in Two
Cohorts of Control Boys (N598) BIC7343.99
Entropy0.789 of Iterations149
Sensitivity
Prevalence of Arrest
34.4
30.0
5.4
FFall SSpring
8
Methodological Topics in Predicting Developmental
Trajectories

• Sensitivity (false positives) versus Specificity
  (false negatives)
• Treatment of Missing Data
• Not only prediction of development, but also
  quality of prediction regarding distal outcomes

9
Time-to-Event Analysis

• Interval Censoring, in addition to left and right
  censoring (i.e., Missing Data problem)
• Analysis of Survival leading to Growth
  development (e.g. timing of school suspension may
  lead to different trajectories in aggression)
• Bivariate Survival analysis (Time to first drug
  use and time to first depressive episode in one
  model)