Dear all,
There is a change of date about the next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine!
It will ultimately take place on Thursday 10 December 2015 at 11h00, in the Salle Delachaux (Biopôle 2, first floor, Route de la Corniche 10, Lausanne, M2 stop: Vennes).
The speaker will be Mark Hannay from the Research Center for Statistics, School of Economics and Management, University of Geneva.
Title: Confidence sets in model selection
Abstract: At first glance the goals of model selection might seem clear. Out of a list of possible models we want to select the "best" or a subset of "best" models. This notion of "best" however is not well defined, since it obviously depends on the initial goals of the selection. In order to study the variability of model selection, we introduce a new definition of a model, where they are no longer defined through exact zeroes components but through negligible components. Then inspired by confidence intervals for estimation of parameters, we propose a method to build confidence sets for model selection in a parametric setting, i.e. create sets of models within which the "true"model is included with a certain confidence. This allows us to perform inference on model selection. We discuss the computational challenges with such a method, how to find p-values (for the model), consistency in model selection and through a data set show the implications of this new method.
Key words: confidence sets of model selection, consistent model selection, variability in model selection
Hope to see you there!
Valentin
(for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz)
CHUV
centre hospitalier universitaire vaudois
Valentin Rousson - Division biostatistique et méthodes quantitatives
IUMSP - Institut universitaire de médecine sociale et préventive
BIO 2/2ème/103
Rte de la Corniche 10, CH - 1010 Lausanne
+41 (0)21 314 7328
valentin.rousson(a)chuv.ch<mailto:valentin.rousson@chuv.ch>
www.iumsp.ch<http://www.iumsp.ch/>
Dear all,
The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 19 November 2015 at 11h00, in the Salle Delachaux (Biopôle 2, first floor, Route de la Corniche 10, Lausanne, M2 stop: Vennes).
The speaker will be Mark Hannay from the Research Center for Statistics, School of Economics and Management, University of Geneva.
Title: Confidence sets in model selection
Abstract: At first glance the goals of model selection might seem clear. Out of a list of possible models we want to select the "best" or a subset of "best" models. This notion of "best" however is not well defined, since it obviously depends on the initial goals of the selection. In order to study the variability of model selection, we introduce a new definition of a model, where they are no longer defined through exact zeroes components but through negligible components. Then inspired by confidence intervals for estimation of parameters, we propose a method to build confidence sets for model selection in a parametric setting, i.e. create sets of models within which the "true"model is included with a certain confidence. This allows us to perform inference on model selection. We discuss the computational challenges with such a method, how to find p-values (for the model), consistency in model selection and through a data set show the implications of this new method.
Key words: confidence sets of model selection, consistent model selection, variability in model selection
Hope to see you there!
Valentin
(for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz)
CHUV
centre hospitalier universitaire vaudois
Valentin Rousson - Division biostatistique et méthodes quantitatives
IUMSP - Institut universitaire de médecine sociale et préventive
BIO 2/2ème/103
Rte de la Corniche 10, CH - 1010 Lausanne
+41 (0)21 314 7328
valentin.rousson(a)chuv.ch
www.iumsp.ch<http://www.iumsp.ch/>
Dear all,
The next seminar in our series of statistics talks for people interested in quantitative methods in biology and medicine will take place on Thursday 12 November 2015 at 11h00, in the Salle Delachaux (Biopôle 2, first floor, Route de la Corniche 10, Lausanne, M2: Vennes).
The speaker will be Prof. Frank Windmeijer (University of Bristol, UK) who will speak about "Selecting (In)Valid Instruments for Instrumental Variables Estimation".
Abstract: Instrumental variables estimation is a procedure for the identification and estimation of causal effects of exposures on outcomes. In Mendelian randomization studies, genetic markers are used as instruments. However, instruments may be invalid when they have a direct effect on the outcome or are associated with the unobserved confounders. We investigate the behaviour of the Lasso for identifying invalid instruments in linear models, as proposed recently by Kang, Zhang, Cai and Small (2015, Journal of the American Statistical Association, in press), where invalid instruments are such that they enter the model as explanatory variables. We show that for this setup, the Lasso may not select all invalid instruments in large samples if they are relatively strong. Consistent selection also depends on the correlation structure of the instruments. We propose an initial estimator that is consistent when less than 50% of the instruments are invalid, but its consistency does not depend on the relative strength of the instruments or their correlation structure. This estimator can therefore be used for adaptive Lasso estimation. We develop an alternative selection method based on Hansen's J-test of overidentifying restrictions, but limiting the number of models to be evaluated. This latter selection procedure is consistent under weaker conditions on the number of invalid instruments and is aligned with the general identification result for this particular model. We apply the methods to estimate the effect of BMI on educational attainment using 95 SNPs as potential instruments.
Hope to see you there!
Valentin
(for the organizers: Jérôme Goudet, Valentin Rousson, Frédéric Schütz)
CHUV
centre hospitalier universitaire vaudois
Valentin Rousson - Division biostatistique et méthodes quantitatives
IUMSP - Institut universitaire de médecine sociale et préventive
BIO 2/2ème/103
Rte de la Corniche 10, CH - 1010 Lausanne
+41 (0)21 314 7328
valentin.rousson(a)chuv.ch
www.iumsp.ch<http://www.iumsp.ch/>
