This 6-part statistics webinar series is annually presented by the Applied Statistics and Data Science Group. Slides and descriptions of each session are provided.
Before analysis begins the data should be visually inspected and explored. We discuss variable types, their appropriate graphics for univariate and multivariate data and provide tips on how to create meaningful and transparent graphics. We will explain how to compute simple data summaries and descriptive statistics and how they help guide any future analysis.
This is a general introduction to the important role statistics play in the planning stage of a research project. We discuss both observational studies and controlled experiments, including the study population, the scientific question, sampling and randomization. We will focus on the experimental design, the effects of confounding, computing sample size and power
A detailed discussion of how to compare data from two groups or conditions using a t-test and the assumptions involved. We generalize to compare more than two groups by analysis of variance (ANOVA) including two-way ANOVA where the groups are defined by two categorical variables.
Model the relationship between two quantitative variables by regression or correlation analysis. We expand the regression model to include additional predictor variables, including other quantitative variables (multiple regression) and categorical variables (ANCOVA) and discuss the importance of checking the model assumptions.
How to compare two groups when the response is binary or count data. What happens if we have more than two groups or more than one predictor whether numeric or categorical. We talk about logistic, Poisson and negative binomial regression analysis with an introduction to generalized linear models and show how to interpret the model coefficients in each case.
Methods like Regression and ANOVA model the mean structure of the observed data while making specific assumptions about the variance. Ignoring the lack of independence due to repeated measurements on the same unit or clusters of related units may lead to wrong conclusions. We show how mixed effects models are used in such cases to model the variance structure of the data as well as the mean.