Predoctoral Fellowships



tel: ++32 9 2644776

Predoctoral Fellowship in statistics: Honest data-adaptive inference for treatment effects


Stijn Vansteelandt,

Department of Applied Mathematics, Computer Science and Statistics, Faculty of Science, Ghent University, Belgium

and Department of Medical Statistics, the London School of Hygiene and Tropical Medicine, U.K.

Tel: +32 (0) 9-2644776

It is known that after variable selection, standard inferential procedures for regression parameters may not be uniformly valid; there is no finite sample size at which a procedure is guaranteed to attain its nominal coverage/size (within pre-specified error margins). This problem is exacerbated in high-dimensional settings where variable selection (e.g. using the Lasso) becomes unavoidable. This has prompted a flurry of activity in developing uniformly valid confidence intervals and hypothesis tests for a low-dimensional regression parameter (e.g. the effect of a treatment A on an outcome Y) in high-dimensional models. Recent proposals aim to ‘debias’ or ‘desparsify’ either the estimating equations for the treatment effect (Avagyan and Vansteelandt 2017; Ning and Liu 2017; Belloni, Chernozhukov, and Hansen 2014) or the estimates themselves (van de Geer et al. 2014).

The aim of this project is to contribute to this literature by developing treatment effect estimators with accompanying confidence intervals (and p-values) that are honest, in the sense of being uniformly valid even when based on model selection strategies or more general data-adaptive methods. The lack of such procedures that acknowledge the uncertainty due to data-adaptive model selection or regularisation constitutes one of the major open problems in statistics. Compared to competitive estimators, the strategies that this project aims to develop

- will be less sensitive to the choice of the model selection strategy or data-adaptive method, as well as to model misspecification;

- will have smaller sampling variability and thus be more precise;

- and the proposed confidence intervals (and p-values) will be straightforward to calculate, whilst also reflecting the uncertainty about the treatment effect after model selection.


The successful candidate will hold an Master degree in Mathematics (or related discipline) or Statistics. She/he will be hosted within a dynamic group of researchers. She/he will be offered excellent training and development opportunities.

Duration: 48 months

Date of start: as soon as possible (negotiable, but no later than August 1)


Avagyan, Vahe, and Stijn Vansteelandt. 2017. “Honest Data-Adaptive Inference for the Average Treatment Effect under Model Misspecification Using Penalised Bias-Reduced Double-Robust Estimation.” ArXiv:1708.03787 [Stat], August.

Belloni, Alexandre, Victor Chernozhukov, and Christian Hansen. 2014. “Inference on Treatment Effects after Selection among High-Dimensional Controls.” The Review of Economic Studies 81 (2):608–50.

Geer, Sara van de, Peter Bühlmann, Ya’acov Ritov, and Ruben Dezeure. 2014. “On Asymptotically Optimal Confidence Regions and Tests for High-Dimensional Models.” The Annals of Statistics 42 (3):1166–1202.

Ning, Yang, and Han Liu. 2017. “A General Theory of Hypothesis Tests and Confidence Regions for Sparse High Dimensional Models.” The Annals of Statistics 45 (1):158–95.

Please send your application (including a current CV and two names of people who are willing to provide a recommendation) to Stijn Vansteelandt. We encourage candidates to apply early. Candidates who are in their final year of study and will graduate before the starting date are also eligible to apply. Applications received before April 1, 2018 will be given full consideration. Applications received after April 1, 2018 will be considered as they arrive, until the position is filled.