Abstract

Various ways to model an integer valued response variable N exist (e.g., zero inflated Poisson, Negative Binomial, etc.). We tackle integer response modeling by simply viewing N as being the integer part of a positive continuous response T. So observing N=8 is akin to observing that continuous T is between 8 and 9 (interval censoring). We are then naturally drawn to survival analysis models. Though several models exists (e.g., Weibull, log-Normal, Cox proportional hazard), we chose to use piecewise constant hazard which is simple, flexible and able to fit adequately. An R package (ModIvIc) was created to assist in model specification and fitting. Using data from the Canadian Community Health Survey (CCHS), this approach was utilised to model weekly alcohol consumption as a function of covariates such as smoking and age.