Regression Models for Heaped Data

• Activity Number: 589
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): James Hardin*+ and Tammy Harris and James Hussey and Alexander McLain
• Companies: University of South Carolina and University of South Carolina and University of South Carolina and University of South Carolina
• Keywords: count data ; heaped data ; overdispersion ; estimation

Abstract

Various approaches to modeling heaped response data are defined and studied. Our initial approach allows the analyst to adjust the specificity of the reported outcomes in a censored data model; separate models for Poisson, negative binomial, and generalized Poisson will be illustrated. Additional models will be presented which simultaneously model the outcome as well as the likelihood of reporting on specified heaping intervals. Motivation for heaped data will be discussed, as well as approaches for comparing heaped models to naive models. Software will be described and illustrated; all software will be available for free download.