Statistics > Methodology
[Submitted on 1 Nov 2015 (this version), latest version 16 Jan 2017 (v3)]
Title:Calibrated Percentile Double Bootstrap For Robust Linear Regression Inference
View PDFAbstract:When the relationship between a response variable $Y$ and covariates $\vec{\mathbf{X}}$ is non-linear with possibly heteroskedastic noise, and where the covariates $\vec{\mathbf{X}}$ are themselves random, the empirical coverage probability of traditional confidence interval (`CI') methods decreases considerably. We propose a double bootstrap-based calibrated percentile method, perc-cal, as a general-purpose CI method which performs very well relative to alternative methods in challenging situations such as these. For the first time, we prove that under relatively mild regularity conditions, the rate of coverage error of perc-cal for a two-sided confidence interval of the best linear approximation between $Y$ and a $p$-dimensional $\vec{\mathbf{X}}$ is $\mathcal{O}(n^{-2})$. We then show that perc-cal performs very well in practice, demonstrating its performance in a thorough, full-factorial design synthetic data study as well as a real data example involving the length of criminal sentences. We have provided an R package, available through CRAN and coded primarily in C++, to make it easier for practitioners to use perc-cal.
Submission history
From: Daniel McCarthy [view email][v1] Sun, 1 Nov 2015 16:49:02 UTC (975 KB)
[v2] Thu, 16 Jun 2016 13:18:44 UTC (459 KB)
[v3] Mon, 16 Jan 2017 17:40:56 UTC (526 KB)
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.