Threshold Value Estimation Using Adaptive Two-Stage Plans in R

• Main Authors: Shawn Mankad, George Michailidis, Moulinath Banerjee
• Format: Article
• Language: English
• Published: Foundation for Open Access Statistics 2015-10-01
• Series: Journal of Statistical Software
• Subjects: threshold estimation, two-stage estimation, R
• Online Access: https://www.jstatsoft.org/index.php/jss/article/view/2376

This paper introduces the R package twostageTE for estimation of an inverse regression function at a given point when one can sample an explanatory covariate at different values and measure the corresponding responses. The package implements a number of nonparametric methods for budget constrained threshold value estimation. Specifically, it contains methods for classical one-stage designs and also adaptive two-stage designs, which have been shown to yield more efficient and accurate results. A major advantage of the methods in package twostageTE is that threshold value estimation is performed without penalization or kernel smoothing, and hence, avoids the well-known problems of choosing the corresponding tuning parameter (regularization, bandwidth). The user can easily perform a two-stage analysis with twostageTE by (i) identifying the second stage sampling region from an initial sample, and (ii) computing various types of confidence intervals to ensure a robust analysis. The package twostageTE is illustrated through simulated examples.