Service interruption on Monday 11 July from 12:30 to 13:00: all the sites of the CCSD (HAL, EpiSciences, SciencesConf, AureHAL) will be inaccessible (network hardware connection).
Skip to Main content Skip to Navigation
Conference papers

Model selection via the lasso in conditional logistic regression

Abstract : We propose a model selection procedure in the context of matched case-control studies and, more specifically, for the conditional logistic regression. The method is based on penalized conditional likelihood with an L1-type penalty of the regression coefficients, the so-called lasso. This penalty, which shrinks coefficients to improve the accuracy of prediction, is particularly adapted when the number of covariates is large (with respect to the number of events) or in case of collinearity between them. An attractive property of the lasso is that it performs parameter estimation and variable selection simultaneously. The implementation of the method is based on a simple modification of the R-package "penalized" for fitting generalized linear and Cox proportional hazard models with L1 or L2 penalties, developed by J. Goeman. The effectiveness of the algorithms allows the use of resampling methods for choosing the regularization term and for evaluating the stability of the selected model. Thus, K-fold likelihood cross-validation and bootstrap are applied, both taking into account the dependent nature of data. The methodology is illustrated with some examples.
Complete list of metadata
Contributor : Marta Avalos Connect in order to contact the contributor
Submitted on : Tuesday, July 7, 2009 - 1:06:22 PM
Last modification on : Monday, April 25, 2022 - 2:10:02 PM
Long-term archiving on: : Tuesday, June 15, 2010 - 5:52:19 PM


Files produced by the author(s)


  • HAL Id : inserm-00402434, version 1



Marta Avalos. Model selection via the lasso in conditional logistic regression. Second International Biometric Society Channel Network Conference, Apr 2009, Ghent, Belgium. pp.Model selection and clustering session. ⟨inserm-00402434⟩



Record views


Files downloads