There Will Be Zeros

Loading Map....

Date(s) - Mar 4, 2008
3:15 PM - 4:15 PM

Center for Community Health


Presented by: 

Scott Comulada, Dr.P.H.

Senior Statistician Jane & Terry Semel Institute for Neuroscience and Human Behavior, and Center for Community Health University of California, Los Angeles


Substance use over time may be impacted by an intervention in two ways. Substance users may stop using, becoming non-users, or users may reduce their use. The complexity of the relationship between an intervention and substance use outcome is often lost if a traditional regression model is fit to the substance use data. This is especially problematic when substance use frequency counts are highly skewed. For example, the frequency of use during the past three months may be mostly zero among the sample of study participants, leading to zero-inflated count data. Intervention efficacy may not be detected if traditional regression techniques are implemented. Another option is to fit a zero-inflated Poisson model that assumes zeros come from two processes. An absence of substance use can result from a user who was not using when queried about their use or a non-user who does not use. A motivating example is provided from a behavioral intervention trial where a reduction in substance use was not detected using traditional techniques and was detected using a zero-inflated Poisson model.


Scott Comulada, Dr.P.H., is a senior statistician for the Center for Community Health at the University of California at Los Angeles (UCLA) with training in biostatistics from UCLA. He has statistical research interests in longitudinal regression models, categorical data analysis and power analysis. He works on several multi-site cognitive behavioral therapy-intervention trials focused on changing HIV-transmission risk behaviors in HIV-positive populations.