Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 59 - Nonparametric Modeling
Type: Contributed
Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Nonparametric Statistics
Abstract #313969
Title: Modeling of Count Data Subject to Measurement Error and Boundary Constraints
Author(s): Cornelis J. Potgieter*
Companies: Texas Christian University
Keywords: deconvolution; errors-in-variables; discrete data

Count variables often have a fixed set of possible outcomes {0,…,N}. Such variables pose a unique problem when measurement error is present in observed data. Specifically, as both the true count X and the noisy count W lie in the fixed set of values, the typical additive model W = X + U with independent measurement error U is unsuitable. While it is possible to extend misclassification models for categorical variables to the count variable framework, the resulting model is tedious and has a large number of parameters to estimate. We explore how a transformation model W = max(0, min(X+U)) can be used to account for measurement error. In this formulation, the additivity and independence of measurement error can still be utilized. The goal of our study is two-fold: Firstly, we propose discrete parametric distributions for the measurement error U that are symmetric and zero mean. Secondly, we consider the companion problems of estimating the pmf of X based on W data and estimating the conditional mean function E[X|W]. The methodology is illustrated in an application involving the assessment of reading abilities of elementary school children.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program