Mission: To review and apply Marginal Structural Methodology for estimating causal parameters in a longitudinal setting. Application of Instrumental Variables methodology to detect and correct chanelling in observational studies.

Means: References:
[1] J.M. Robins. Marginal structural models versus structural nested models as tools for causal inference. In Statistical models in epidemiology, the environment, and clinical trials (Minneapolis, MN, 1997), pages 95–133. Springer, New York, 2000a.

[2] M.J. van der Laan and J.M. Robins. Unified methods for censored longitudinal data and causality. Springer,New York, 2002.

Programming language: SAS, R

Mission: To compare two models of the intra-genes variance. The models were structural models and both suggested to shrink the variance using a bayesian prior. My work consisted first in a comparison of both methods based on the sensitivity, the specificity, the false positive and false negative rates; then in classifying the genes differentially expressed so as to appreciate their nearness. I was involved in a team made of four colleagues and supervised by Dr F. Jaffrezic, statistician at the INRA (National Institute of Research in Agronomy).

Means: Mixed models, Bayesian statistics backgrounds, R programming

Mission: To assess the metrological properties of a quality of life scale, namely the Functionnal Assessment of Cancer Therapy-General (FACT-G). The ability of the FACT-G scale to measure quality of life of cancer patients have been already desmontrated in previous studies. This additionnal study aimed at comparing the coherency of the French translation of the questionnaire. I was involved in a team made of four colleagues and supervised by N. Costet, statistician at the INSERM (National Institute of Health and Medical Research).

Means: General needs: ability to work in team efficiently.
Technical needs: ANOVA, M-Estimation, Tests Theory, and more generally backgrounds of inferential statistics.

Appraisal: The conclusion of our study have been consistent with the previous assessments that have been already published. We have been able to bring to light some typical problems due to the psychometrical measures such as the response shift effect. Ours conclusions have then contributed to broaden the use of the this scale. The whole report is available.