Transcript Synthesis and general discussion of data

INSERM Workshop, St. Raphael
Synthesis
(personal point of view…)
Bruno Falissard
Univ. Paris-Sud, INSERM U669
Synthesis
• The objective of a statistician is
– 1/ To understand, formalize a question and to
propose an appropriate answer from the analyze
of a dataset
– 2/ To be confident in this answer
– 3/ As possible, to use analyses that are
understandable by the potential reader
Synthesis
• Mixture models for longitudinal data are
fecund (Sylvana)
• They can be used to obtain trajectories
• They can be used for more
Obtaining trajectories
• K-means algorithm (KmL)
– “Non parametric”
– First intention
– Exploratory approach
– Good pair of glasses
– 1/, 2/, 3/
Obtaining trajectories
• Proc Traj
– Statistical Model
– Trajectories likely to be of a better quality
– 1/, 2/, 3/
Obtaining trajectories
• Questions
– What is a trajectory? (realism versus nominalism)
– Group membership is a convenient statistical
fiction
– The question of the number of trajectory is
quixotic
– Valorization of trajectories: crosstab
Obtaining more than trajectories
•
•
•
•
Daniel (PS)
Cécile (mixed models with LC)
Hélène (joint models with LC)
Bengt (drug only responders, non ignorable
drop outs)
• Thiomir (survival)
• Maria (bayesian, piecewise parametric)
• José (classification)
Obtaining more than trajectories
• Sophisticated models integrating two
approaches
– Fully explicit
– Bottom-up / top-down
– P-values and BIC
Obtaining more than trajectories
• Perhaps two different paradigms
– Local: Hélène/Cécile
– Global: M-plus
– Remark of J.L. Foulley
• Mplus is a new language?
– 1/, 2/, 3/ ?
• Obviously, there is here a promising
and emerging field:
–Of applications
–Of research