Learning Objectives
- Define and identify over-dispersion in count data
- Define the negative binomial (NB) distribution and identify applications for it
- Define zero-inflated count models
- Fit and interpret Poisson and NB, with and without zero inflation
Outline
- Review of log-linear Poisson glm
- Review of diagnostics and interpretation of coefficients
- Over-dispersion
- Negative Binomial distribution
- Zero-inflated models
- Vittinghoff section 8.1-8.3
Learning Objectives
- Fit Poisson, NB, and zero-inflated loglinear models
- Perform nested deviance test for model selection
- Make diagnostic plots of loglinear models
Exercises
-
Load the needle-sharing dataset.
- Compare the mean to the variance of the outcome variable.
- Calculate what fraction of the counts are zero.
- Create a histogram of the outcome variable.
-
Fit Poisson and Negative Binomial models as in the lecture, with and without zero inflation.
-
Use chi-square nested deviance tests to assess which model seems to fit best.
-
Create residual deviance plots using the functions defined in the lecture.
-
Plot the frequences of predicted and observed counts for each model.