Transcript General introduction - Christophe Genolini

INSERM Workshop, St. Raphael
Mixture modeling for
longitudinal data
Introduction
Bruno Falissard
Univ. Paris-Sud, INSERM U669
Mixture modeling for longitudinal data
• Everybody says that longitudinal data are
essential
• Most often
– T1, T2, T3, T4
– T4 explained by T1
• Another perspective : typical patterns of
evolution across time
Mixture modeling for longitudinal data
Temperature
People with fever receiving an antibiotic
time
Mixture modeling for longitudinal data
People with fever receiving an antibiotic
Temperature
39
37
time
Mixture modeling for longitudinal data
Temperature
People with fever receiving an antibiotic
time
Mixture modeling for longitudinal data
People with fever receiving an antibiotic
Temperature
Virus
Bacteria
time
Mixture modeling for longitudinal data
•  good or bad responders in RCTs
•  developmental perspective
• …
Mixture modeling for longitudinal data
• Why this workshop today and not several
years ago?
– The questions did exist several years ago
– But
• Somewhat exploratory approach (not classical in
biomedical research)
• Tools more or less efficient (big sample sizes,
transversal data)
Mixture modeling for longitudinal data
• But the first applications appeared…
– Trajectory of aggression in young children
• With user-friendly routines
• With some elements of robustness
Mixture modeling for longitudinal data
• These models are somewhat different from
the classical techniques used in biomedical
research
– “Bottom up” as opposed to “top down”
– Intersection of numerous methodological fields
•
•
•
•
Biostatistics
Computer science
Social sciences
Psychometrics
Mixture modeling for longitudinal data
• Statistics are not only mathematics, statistics
are highly dependant on the background of
application (culture)
– Unique opportunity to confront different type of
approaches
– With statistical and practical considerations
– With the objective to be confident with these
methods, and to be able to explain them to
reviewers