Keynote Presentation

Real-world questions about using regression models for counts

Real-world data and research questions have always been the motivation for applied statistical research. Many research questions in psychology and education involve counts of behaviors or events, such as the number of depressive symptoms a person has experienced in the past month. These variables have unique characteristics – discrete, non-negative, and right-skewed – that make them inappropriate to serve as outcomes in standard analyses such as ANOVA and linear regression. Poisson regression and related models like negative binomial regression are the preferred methods for count outcomes. Poisson regression has been extended for more broad purposes, such as to allow modeling of “excess” zeroes in counts and longitudinal models for count outcomes. My current research involves extending count models to appropriately predict count outcomes that have been coarsely-grouped and to construct measures of effect size that can be better compared with linear model effect sizes such as Cohen’s d.

Date: Thursday, January 19
Time: 11:00 a.m. to 12:00 p.m.
Location: Nebraska Union, Auditorium

The keynote is free, open to the public and requires no registration.

Statistics Workshop

Generalized linear models: All in the family

The general linear model (GLM), which includes linear regression and ANOVA, is a widely-used analytic framework for psychology and many other fields. The GLM assumes that the outcome variable is continuous, conditionally normally distributed, and has constant variance. Many research questions involve bounded, non-continuous, or non-normally distributed outcome variables that do not satisfy these assumptions. The generalized linear model (GLiM) is a large, flexible family of models that extends the GLM framework to accommodate non-normal outcomes. Such outcomes include binary variables, ordered categories, counts, proportions, and many others. Researchers are often familiar with individual models for non-normal outcomes – such as logistic regression for binary outcomes – but are not aware of the larger family in which these models exist. This workshop will discuss the theoretical basis of the GLiM family and its individual models (part 1), and applications of several commonly used GLiMs (part 2).

Part 1
Date: Thursday, January 19
Time: 3:00 to 5:00 p.m.
Location: Nebraska Union, Heritage Room

Part 2
Date: Friday, January 20
Time: 10:00 a.m. to 12:00 p.m.
Location: Nebraska Union, Auditorium

The workshop is free but requires registration. Seating is limited.

Register Now