Friday, February 24
CS10 Probability Distributions Fri, Feb 24, 2:00 PM - 3:30 PM
River Terrace 3

Modeling Proportions and Probabilities: The Beta Distribution Is Your Friend (303366)

*Paul Teetor, William Blair & Co. 

Keywords: proportions, beta distribution, beta regression, beta-binomial model

Applied statisticians often encounter proportions, probabilities, and other unit-interval data, such as failure rates and transition probabilities. It’s natural to model that data using a beta or binomial distribution, but the relevant techniques are not widely known and there are pitfalls along the way. How do we build a regression model of a beta variable? Can we adapt the binomial model when the data is over-dispersed? Is the full complexity of Bayesian modeling necessary?

This presentation will explain and demonstrate fitting beta and beta-binomial models in practical applications. Topics will include

- Review of the beta distribution - Fitting the beta distribution - Introduction to beta regression - Fitting the beta regression - Limitations of the binomial distribution, especially for over-dispersed data - Advantages of the beta-binomial model - Fitting the beta-binomial model

Examples will use real data sets, including sports data and a stock-ranking application. All concepts will be illustrated with R code and graphical plots.