Effects of Ignoring Truncation in Poisson Count Models

• Activity Number: 28
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Abdalhalim Suaiee*+ and Trent L. Lalonde
• Companies: University of Zawia and University of Northern Colorado
• Keywords: Double-Truncation ; Model Misspecification ; Poisson Regression ; Truncated Counts

Abstract

Count response data situations arise often in practice, and are typically modeled using a generalized linear model with a Poisson response distribution. A Poisson distribution imposes the assumption that the data are observed on the interval of all non-negative integers. However, practical applications often involve restrictions that reduce the domain of possible response values. Such data are referred to as ``truncated" count responses. While zero-truncated data are well-recognized and often accounted for, little attention has been paid to general left-truncation, right-truncation, or double-truncation of counts. It is useful to be aware of the consequences of misspecification of a model such that truncation of any type is ignored. In this paper we compare model performance when truncation is not accounted for, when truncation is partially accounted for, and when truncation is completely accounted for. All three cases of left-truncation, right-truncation, and double-truncation are investigated.