Generalized linear models are an extension of ordinary linear regression models when the response is not normally distributed. This is the case when working with count data, as counts are whole numbers, bounded by zero and often have a right skewed distribution. The workshop will first introduce an appropriate discrete distribution for count data: the Poisson distribution. Then the application and interpretation of the type of generalized linear models most often used for count data, namely, Poisson regression models, will be covered. The workshop will also discuss how to detect and mitigate the effects of overdispersion using quasi-poisson and negative binominal regression models. Zero-inflated models and hurdle models will be discussed as techniques to deal with high occurrence of zeros in the observed data.
After attending this workshop, you should be able to:
- Decide whether Poisson regression is appropriate for your data
- Implement and interpret results from a Poisson regression analysis
- Test which Poisson regression model best fits the data
- Detect and understand the effects of overdispersion and zero-inflation
- Determine whether a hurdle model or a zero-inflated model is applicable