Guillermo Basulto-Elias, Computation of kernel deconvolution density estimators

Friday, August 5, 2016 -
11:00am to 11:45am
Event Type: 

In many areas of application, like medical sciences, variables of interest are not directly observable and may be measured only in the presence of contaminating errors. These cases are often referred as ``measurement error problems.'' Kernel deconvolution density estimation (KDDE) is an approach for handling such measurement errors, which consists of separating out and estimating the density of a target variable from observations blurred by additive errors.  The method involves an adaptation of kernel density estimation using the Fourier inversion theorem. The resulting estimator requires numerical integration of complicated functions. We will talk about how to efficiently perform KDDE in R under several sampling scenarios for univariate and bivariate samples. Several approximations to the measurement error are allowed in estimation, depending on the nature of data collected and amount of knowledge presumed about the underlying error distribution.