Lalonde, Trent

Committee Member

Holighi, Khalil Shafie

Committee Member

Schaffer, Jay R.

Committee Member

Ku, Heng-Yu


Department of Applied Statistics and Research Methods


University of Northern Colorado

Type of Resources


Place of Publication

Greeley (Colo.)


University of Northern Colorado

Date Created



175 pages

Digital Origin

Born digital


Count regression models are used when the response variable takes count or non-negative values. Poisson and negative binomial distributions are commonly used to model count data. A frequent matter with the count data is to have an excess number of zeros that can result in overdispersed data when using Poisson or negative binomial distributions. Appropriate approaches to use when modeling excess-zero data is to use either a hurdle or a zero-inated Poisson (ZIP) distribution. Recently, the hurdle models are commonly used in fields such as medicine, epidemiology, genetics, and marketing. Excess-zero data occur frequently as a series of data that are repeatedly measured over time as well. In this dissertation, the hurdle distribution is used to model time series data that are counts with a high frequency of zeros. Particularly, a first order autoregressive hurdle process is formulated to model excess-zero time series data. Comparisons with two existing zero-inate time series models are presented and the models are evaluated based on their prediction capabilities. It is concluded that the developed hurdle autoregressive model provides better prediction of future observations compared to the other zero-inated Poisson models. The three models are used to analyze the crime data and the results show that the three models do not provide good prediction of future observations.

Degree type


Degree Name


Local Identifiers


Rights Statement

Copyright is held by the author