Post‐selection point and interval estimation of signal sizes in Gaussian samples

Abstract

We tackle the problem of the estimation of a vector of underlying means (signal sizes) from a single vector‐valued observation y. Often one is interested in estimating only a subvector of signals corresponding to a set of selected, “interesting” sample elements. These “interesting” sample elements tend to have the largest absolute size, gleaned by applying some selection procedure like that of Benjamini & Hochberg (2015). Previous work on this estimation task proposes the reduction in size of the largest (absolute) sample elements either via shrinkage (like James–Stein) or by subtracting biases estimated using empirical Bayes methodology. We take a novel approach and adapt recent developments by Lee et al. (2016) in post‐selection inference. Adapting and applying their distributional results to our problem post‐selection point and interval estimators for underlying signal sizes are proposed. Simulations suggest that our estimator seems to perform quite well against competitors. Furthermore we prove an upper bound to the so‐called “worst case risk” of our estimator—when combined with the Benjamini–Hochberg selection procedure—and show that it is within a constant multiple of the minimax risk over a rich set of parameter spaces meant to evoke sparsity. The Canadian Journal of Statistics 45: 128–148; 2017 © 2017 Statistical Society of Canada

Low-Complexity Methods for Estimation After Parameter Selection

“…In post-selection inference, it is well known that the selection stage has an impact on subsequent estimation, by creating coupling between parameters that originally were decoupled [14], leading to inaccurate confidence intervals, and introducing a selection bias [15]- [21]. The enlightening example by Efron [20, Fig.…”

“…The PSMSE, which is the mean squared error (MSE) of the selected parameter, is widely used in the mathematical statistics literature and in practical experiment design [21]- [27]. We and others (see e.g.…”

Adjusting for selection bias in assessing treatment effect estimates from multiple subgroups

“…Later, Cohen and Sackrowitz (1989), Stallard and Todd (2003), Bowden and Glimm (2008), and Stallard and Kimani (2018) focused on unbiased point estimation for selected means. Regarding applications in genomics, several authors have focused on the case of huge numbers of subgroups (potentially in the millions) (Efron, 2010;Reid, Taylor, & Tibshirani, 2017). Regarding applications in genomics, several authors have focused on the case of huge numbers of subgroups (potentially in the millions) (Efron, 2010;Reid, Taylor, & Tibshirani, 2017).…”

“…Venter (1988), Bebu, Luta, and Dragalin (2010), Bebu, Dragalin, and Luta (2013), Fuentes, Casella, and Wells (2018) also consider interval estimation of selected effects. Regarding applications in genomics, several authors have focused on the case of huge numbers of subgroups (potentially in the millions) (Efron, 2010;Reid, Taylor, & Tibshirani, 2017). In this case, it is commonly assumes that only a very small subset of all subgroups is relevant.…”

Selective peak inference: Unbiased estimation of raw and standardized effect size at local maxima

“…Efron (2011) uses an empirical Bayes technique to correct for this bias, an approach that has been applied in the genetics literature (Ferguson et al (2013)). In the case of independent random variables that each come from distributions belonging to a known parametric family, Simon and Simon (2013) introduced a frequentist method to correct bias and Reid et al (2014) details a post-model selection approach which involves calculating the distribution of Gaussian random variables conditional on being selected.…”